THE ROLE OF CONTEXTUAL INFORMATION IN EXPERT ANTICIPATION

Transcription

1 THE ROLE OF CONTEXTUAL INFORMATION IN EXPERT ANTICIPATION COLM PÁDRAIG MURPHY A thesis submitted for the degree of Doctor of Philosophy Department of Life Sciences, Brunel University London July 2017

2 Contents Acknowledgements... 3 List of Publications and Presentations... 4 List of Figures... 5 Abstract... 7 Chapter 1: Perceptual-cognitive expertise and an introduction to the role of contextual information in anticipation... 8 Chapter 2: Contextual information and perceptual-cognitive expertise in a dynamic, temporally-constrained task Chapter 3: The sources of contextual information contributing to skilled anticipation Chapter 4: Informational constraints, option generation and anticipation Chapter 5: Epilogue References Appendix

3 Acknowledgements I could not possibly attempt to mention here, every single person who has helped me over the past few years. I can only say, with the greatest sincerity, to everyone who has played a role, no matter how big or small that role may have been, thank you. However, I would particularly like to express my gratitude to some key people, without whom this would not have been possible. I count myself incredibly lucky to have been supervised by true experts in the field and I appreciate every minute they have spent guiding me through the PhD process. I would like to thank Professor Mark Williams for taking everything he knows about skill acquisition and applying it to PhD supervision with the greatest efficiency, for providing me with endless opportunities, priceless advice and perfectly timed encouragement. Dr Robin Jackson for his equally priceless advice, invaluable guidance and impressive patience. I am sure at times he cursed his open door policy but never once showed it. Dr Dan Bishop, for his support and assistance on the (relatively long) home straight. To those who facilitated data collection and everyone who acted as a participant, I cannot express my gratitude enough for the time and energy they committed. Equally as importantly, there are certain people in my personal life who I would like to thank for supporting me throughout this long journey. Melissa, for being with me every single step of the way. We have been on many adventures together but without her, this one would not have been remotely possible. Her persistent encouragement has been awesome. My parents, for providing me with the tools to choose the path I wanted to. All my family, for their unwavering support and abundant understanding. Finally, my friends, for being exactly that. Thanks everyone. 3

4 List of Publications and Presentations Publications Murphy, C. P., Jackson, R. C., & Williams, A. M. (2018). The role of contextual information during skilled anticipation. Quarterly Journal of Experimental Psychology. Murphy, C. P., Jackson, R. C., Cooke, K., Roca, A., Benguigui, N., & Williams, A. M. (2016). Contextual information and perceptual-cognitive expertise in a dynamic, temporally-constrained task. Journal of Experimental Psychology: Applied, 22, Conference Proceedings Murphy, C. P., Jackson, R. C., & Williams, A. M. (2016). The effect of sequencing information on anticipation. Journal of Sport and Exercise Psychology, 38, S90. Murphy, C. P., Jackson, R. C., & Williams, A. M. (2015). Perceiving context: The key to anticipation in sport? In O. Schmid, S. Seiler (Eds.) 14th European Congress of Sport Psychology Bern, Switzerland, 154. Murphy, C. P., Jackson, R. C., Roca, A., & Williams, A. M. (2015). Cognitive processes underlying anticipation in a context-oriented task. Journal of Sport and Exercise Psychology, 37, S53. Murphy, C. P., Jackson, R. C., & Williams, A. M. (2014). The use of contextual information in expert tennis anticipation. In A. De Haan, C. J. De Ruiter, E. Tsolakidis (Eds.) 19th Annual Congress of the European College of Sport Science, Amsterdam, The Netherlands. 4

5 List of Figures 1.1 The three stages of Ericsson and Smith s (1991) expert performance approach Percentage of correct anticipatory movements as a function of response initiation time relative to the opponent s racket-ball contact in professional tennis matches (t = 0) Video (top) and animated (bottom) display conditions Mean (SE) depth, direction and combined response accuracy of skilled and lessskilled participants in video and animated display conditions. *p < Mean (SE) depth, direction and combined response accuracy across groups in the animated display condition. *p < Mean (SE) number of verbal statements made by skilled and less-skilled participants. *p < Mean (SE) percentage of reports containing keywords for skilled and less-skilled participants. *p < Mean (SE) percentage viewing time relative to fixation location of skilled and less-skilled participants. *p < Representation of the number of shots played in each Sequence Length condition prior to the same shot occluded at the opponent s racket-ball contact Mean (SE) depth (a), direction (b) and combined (c) response accuracy relative to sequence length across groups. *p <

6 3.3 Mean (SE) depth (a), direction (b) and combined (c) response accuracy in the six display and presentation conditions across groups. *p < Mean (SE) confidence ratings of skilled and less-skilled participants in the six display and presentation conditions. *p < Mean (SE) depth, direction and combined response accuracy across groups in video and animated display conditions. *p < Mean (SE) number of task-relevant and task-irrelevant options generated per trial by skilled and less-skilled participants in video and animated display conditions. *p < The relationship between mean number of options generated and combined response accuracy across groups in video (a) and animated (b) display conditions The relationship between mean number of task-relevant and task-irrelevant options generated and combined response accuracy in video (a) and animated (b) display conditions

7 Abstract While it is well established that expert performers can pick up and utilise postural cues to anticipate more effectively than less-skilled counterparts, the role of contextual information in expert anticipation has received relatively little research attention. The aims of this thesis were to highlight the importance of contextual information in anticipation, identify specific sources of contextual information that impact anticipation, and examine how this information is used. In five studies, skilled and less-skilled tennis players were presented with videos or animations of the same open play rallies. The animations omitted postural information, constraining participants to anticipate based on contextual information alone. First, participants anticipated more accurately than chance in both display conditions. Skilled participants were more accurate than less-skilled participants, with the difference being greater in the video condition. Second, gaze data and retrospective verbal reports were collected when viewing the animations. Skilled participants displayed different gaze behaviour and more thoroughly evaluated the presented information than less-skilled participants. Third, animations were manipulated to depict or omit potential sources of contextual information. The preceding shot sequence was shown to be a useful source of contextual information, particularly for skilled participants. Additionally, player positioning could be used to anticipate highly accurately in absence of any other information. Finally, the option generation strategies underpinning expert anticipation were examined. Participants generated fewer options when postural cues were available compared with when constrained to the use of contextual information alone. Moreover, skilled participants generated more task-relevant and fewer task-irrelevant options than less-skilled participants. Collectively, these findings increase understanding of the role of contextual information in expert anticipation and further highlight the complex nature of perceptual-cognitive expertise. 7

8 Chapter 1 Perceptual-cognitive expertise and an introduction to the role of contextual information in anticipation 8

9 It is difficult not to be fascinated by the amazing behaviour and feats of expert performers. There is perhaps no other domain in which the superiority of experts is more easily apparent than sport. At the highest levels, expert athletes can not only execute actions with phenomenal speed and accuracy, but they can respond to the equally quick and accurate actions of their opponents whilst giving the impression of having all the time in the world. The competitive nature of sport means that it is rife with individuals striving for excellence and constantly pushing the boundaries of human performance in an attempt to become the best. It therefore provides a particularly fruitful domain within which to conduct research that allows us to move beyond the realms of fascination, towards a fuller understanding of the essential skills and attributes of expert performance as well as the underlying processes and mechanisms (Williams, Ford, Eccles, & Ward, 2011). Knowledge gained from this field of research can be used to develop and adapt existing theoretical models of skill acquisition and expertise, both in sport and in domains such as medicine and education (Farrow, Reid, Buszard, & Kovalchik, 2017; Williams & Ericsson, 2005), as well as providing important implications for training interventions and talent identification and development programmes (Ericsson & Ward, 2007; Williams & Reilly, 2000). For over a century, there has been debate about what distinguishes expert performers from their less-expert counterparts. The early research findings of Sir Francis Galton (1869) suggested that expertise is determined by an individual s innate ability and motivation, a view which was widely held for many years with few exceptions (e.g., Watson, 1924). Given the perceived importance of an individual s genetic make-up in determining expertise, researchers invested great effort in identifying and describing the innate characteristics and abilities underlying skilled performance (e.g., Ackerman, 1988; Fleishman, 1972; Terman & Oden, 1959). 9

10 Early research findings suggested that expert sports performance was underpinned by superior visual attributes (e.g., Fullerton, 1921). However, more recent research has led to a general consensus that expert sports performance is not limited by basic visual functions such as visual acuity and depth perception (e.g., Abernethy, 1987; Abernethy, Neal, & Koning, 1994; Helsen & Starkes, 1999; Starkes & Deakin, 1984; Ward & Williams, 2003). Many of these researchers have employed multivariate analyses to determine the degree of the variance between expert and less-expert performers which can be accounted for by tests of vision, perception and cognition. Rather than measures of visual function, it is measures of perceptual-cognitive skills such as anticipation and decision making that consistently account for a large proportion of the variance (e.g., Abernethy et al., 1994; Ward & Williams, 2003). Mann and colleagues recently presented further evidence against the notion that superior vision is a limiting factor of expert performance (Mann, Ho, De Souza, Watson, & Taylor, 2007; Mann, Abernethy, & Farrow, 2010a, 2010b; see also Applegate & Applegate, 1992). Using a visual blurring technique to degrade the visual clarity of skilled cricket batters, Mann et al. demonstrated that substantial decrements in visual clarity, in some cases to a level comparable to legal blindness, were necessary for batting performance to be negatively affected. Expertise It was through the seminal research of de Groot (1965) and later, that of Chase and Simon (1973) that some of the earliest evidence emerged to suggest that expertise results from the development of domain-specific skills rather than innate individual differences. De Groot (1965) presented chess players of various levels with structured configurations of chess pieces for a short period of two to 10 seconds. After having withdrawn these pieces from sight, participants were asked to recall the positions of the pieces from memory. Chess Grand Masters and Masters were able to reproduce the positions of these 10

11 pieces with approximately 93% accuracy. However, in the three progressively lower classified skill levels, recall accuracy drastically decreased (approximately 72%, 51% and 33%). Chase and Simon (1973) extended this research by asking participants to recall structured configurations of pieces extracted from games between advanced chess players, as well as configurations of pieces which had been randomly arranged on the board. When structured configurations were presented, their findings were comparable with those of de Groot (1965). However, of particular significance is that when required to recall the pieces presented in random configurations which would not usually be observed in a chess match, the Master was no more accurate than the A level or Club level player. It could therefore be concluded that the Master s superior performance when recalling the structured configurations was not due to greater visual short term memory (STM) capacity (which would have been evidenced by more accurate recall regardless of the structure of the configurations). The differences in recall in the structured condition alone must have been due to the context provided by the positions of the chess pieces. Viewing these familiar configurations which the Master would have frequently experienced and engaged with previously facilitated superior recall, this process made possible by a more extensive, domain-specific knowledge base. Chase and Simon (1973) proposed a chunking mechanism to explain the superior recall of expert chess players. They suggested that when the configurations were structured, the more expert chess players were able to perceive this structure and encode large numbers of pieces (15-30) into meaningful chunks, rather than needing to encode pieces one by one, which would soon exceed the capacity of STM (7 ± 2 items; Miller, 1956). While chunking increases the amount of information which could be encoded to circumvent the limitations of STM, the proposed mechanism nevertheless functions 11

12 within the limitations of STM in that the number of chunks, cannot exceed the capacity of STM. Since the seminal work of de Groot (1965) and Chase and Simon (1973), the ability of expert performers to encode and recall domain-specific patterns has been observed in multiple sports such as basketball (Allard, Graham, & Paarsalu, 1980; Gorman, Abernethy, & Farrow, 2012) field hockey (Starkes & Deakin, 1984), football (Williams, Davids, Burwitz, & Williams, 1993; van Maarseveen, Oudejans, & Savelsbergh, 2015) and handball (Tenenbaum, Levy-Kolker, Bar-Eli, & Weinberg, 1994), as well as in other domains such as computer programming (Adelson, 1981), electronics (Egan & Schwartz, 1979) and map reading (Gilhooly, Wood, Kinnear, & Green, 1988). Although Simon and Chase (1973) acknowledge that their research did not directly address how expertise is developed, they did suggest a minimum of ten years practice in chess would be necessary to become a Grand Master. Other researchers in later years echoed this proposal, agreeing that extensive preparation of approximately ten years or more is a common requirement for attaining expertise in many sports and in other domains (e.g., Bloom, 1985; Ericsson & Crutcher, 1990). Ericsson, Krampe, and Tesch- Romer (1993) reiterated this point in their deliberate practice framework. They suggested that through engagement in large amounts of domain-specific practice, it is possible to develop the cognitive skills necessary to circumvent the limits of STM capacity, which facilitate expert performance. Ericsson and Smith (1991) criticised early approaches as attempts to independently measure constructs that were thought to be factors underpinning expertise, rather than describing and subsequently analysing the components that contribute to expert performance in standardised conditions. The pioneering work of de Groot, Chase, and Simon provided the foundations for Ericsson and Smith to outline the expert performance 12

13 approach, a three-stage systematic framework for empirically examining expertise (see Figure 1.1). Figure 1.1. The three stages of Ericsson and Smith s (1991) expert performance approach 1. Taking this approach, researchers first design tasks to capture expert performance in the laboratory or field. Advances in technology have made this increasingly possible with expert performance now consistently being captured using standardised tasks in laboratory settings (e.g., McRobert et al., 2013; Spitz, Put, Wagemans, Williams, & Helsen, 2016). Second, the processes and mechanisms underlying expert performance are identified. Tasks manipulations (e.g., film occlusion, displaying information in point-light format) and process-tracing methods (e.g., gaze tracking, verbal report analysis) are commonly employed at this stage (e.g., Abernethy, Zawi, & Jackson, 2008; McRobert, Williams, Ward, & Eccles, 2009). Finally, the key acquisition processes associated with expertise development are investigated. Practice history profiling and training interventions have proven instrumental in providing evidence of the practice activities 1 Reprinted from Human Movement Science, 24, Williams, A. M., & Ericsson, K. A., Perceptualcognitive expertise in sport: Some considerations when applying the expert performance approach, p. 286,

14 and environmental factors that contribute to expert performance (e.g., Broadbent, Causer, Ford, & Williams, 2015; Roca, Williams, & Ford, 2012). Perceptual-cognitive expertise Perceptual-cognitive skill is an individual s ability to identify and acquire environmental information for integration with existing knowledge such that an appropriate response can be selected and executed (Marteniuk, 1976). It is now widely acknowledged that some of the most important adaptations resulting from extended experience and practice in a particular domain are in the perceptual and cognitive skills underpinning performance (Baker & Farrow, 2015; Williams & Ford, 2008). This is not to discount the importance of physiological components of performance, but rather to note that perceptual-cognitive skills such as anticipation and decision making have been demonstrated to account for a greater proportion of the variance between the performance of experts and less-expert counterparts (Williams & Reilly, 2000). While the importance of perceptual-cognitive skills in domains such as military combat (Williams, Ericsson, Ward, & Eccles, 2008), medicine (Krupinski, 2000; McRobert et al., 2013), driving (McKenna & Horswill, 1999; Underwood, 2007), and law enforcement (Ward, Suss, Eccles, Williams, & Harris., 2011) is indisputable, there are perhaps few domains in which the importance of such skills is as prominent or as obvious as in sport. In tennis, for example, service speeds can reach over 200 km h -1 (Mecheri, Rioult, Mantel, Kauffmann, & Benguigui, 2016) with the distance between the two players being roughly 24 m at the moment the ball is struck. It can therefore take less than 500 ms from the moment the ball leaves the opponent s racket to reach, and potentially pass, the receiving player (Abernethy & Wollstein, 1989). Similarly tight time constraints can be observed in other fast ball sports such as cricket (Regan, 1997), baseball (Gray, 14

15 2002a), and football (Dicks, Button, & Davids, 2010). Such constraints provide researchers with an ideal domain in which to investigate how expert performers circumvent information processing limitations to perform effectively (Moran, 2009). It is generally accepted that, due to extensive experience and practice in their domain, expert performers develop more extensive knowledge bases than less-expert performers, and that these knowledge bases allow experts to more effectively attend to and interpret relevant information in the environment to anticipate upcoming events (Mann, Williams, Ward, & Janelle, 2007). In the following sections, descriptions of some of the specific perceptual-cognitive skills proposed to underpin expert anticipation are presented. Postural cue utilisation The ability to pick up and utilise relevant information from an opponent s bodily movements and postural orientation to anticipate their intentions in advance of a critical event (e.g., racket-ball contact in tennis) has proven to be a reliable discriminator between expert and less-expert performers (e.g., Causer, Smeeton, & Williams, 2017; Goulet, Bard, & Fleury, 1989; Moore & Müller, 2014). Researchers originally demonstrated this skill through expert-novice comparisons on tasks employing the temporal occlusion paradigm. Employing this method, researchers usually present participants with video footage of the opponent executing an action. These videos are edited to selectively occlude at various time points prior to, at, or following the critical event. Using this approach, Jones and Miles (1978) demonstrated that tennis players could anticipate the direction of an opponent s serve in advance of ball flight information becoming available and that skilled players were more accurate than less-skilled counterparts when anticipating the opponent s intentions prior to racket-ball contact. The expert advantage in picking up and utilising postural cues to anticipate the opponent s intentions has since been reproduced in a variety of sports (e.g., Abernethy & Russell, 1987; Rosalie & 15

16 Müller, 2013; Starkes, 1987; Williams, 2000). Researchers employing this paradigm have further demonstrated that observed skill differences are more pronounced when vision is occluded earlier in the opponent s movement (e.g., Müller, Abernethy, & Farrow, 2006; Tenenbaum, Levy-Kolker, Sade, Lieberman, & Lidor, 1996; Williams & Burwitz, 1993), suggesting that experts develop the ability to pick up specific early-occurring cues to overcome information processing limitations in situations of extreme time constraints. While the temporal occlusion paradigm has been employed to determine the time point at which critical information becomes available, the spatial occlusion paradigm has been employed to determine the specific postural cues used to anticipate the opponent s intentions. Using this approach, researchers selectively occlude certain areas of the visual display or alternatively present only certain areas for the duration of the action up until the moment of the critical event (Abernethy & Russell, 1987; Hagemann, Schorer, Cañal- Bruland, Lotz, & Strauss, 2010; Jackson & Mogan, 2007). If performance drops significantly when a certain area of the body or equipment is occluded from the visual display (in comparison to a control condition without occlusion), it can be assumed that this area provides pertinent postural cues to the performer. More recently, researchers have combined temporal and spatial occlusion methods to determine the time point at which specific cues are extracted in anticipation (Causer et al., 2017; Müller et al., 2010). Although the information sources relied upon vary between sports, experts have generally been shown to pick up earlier-arising cues from more proximal information sources (e.g., the hips in football [Causer et al., 2017] and the trunk in fencing [Hagemann et al., 2010]) than less-skilled athletes who rely more on later-occurring, more distal information. Based on research examining biological motion perception (e.g., Johansson, 1973, Kozlowski, & Cutting, 1977; Shim & Carlton, 1997), employment of the point-light method has been instrumental in determining the nature of the information which 16

17 facilitates anticipation. This method involves presenting participants with point-light displays (usually white dots of light on a black background or vice-versa) of an opponent executing an action in which the depicted information has been reduced to minimal biological motion information. Although a slight drop in performance is usually observed when viewing point-light displays in comparison to videos, the skill differences observed in studies employing the temporal and spatial occlusion paradigms are generally replicated when point-light displays are presented rather than videos (e.g., Abernethy, Gill, Parks, & Packer, 2001; Abernethy & Zawi, 2007; Abernethy et al., 2008). Several researchers have interpreted this finding as meaning that the information facilitating expert anticipation is picked up from the relative motion between specific body parts or equipment rather than isolated body parts or superficial features. Experts have been shown to more effectively and efficiently scan and subsequently fixate on more informative areas of the visual display to pick up pertinent postural cues when anticipating an opponent s intentions (Mann et al., 2007). This effect has frequently been demonstrated via skill-based differences in the number of locations fixated on, the duration or length of these fixations, or the time spent fixating on certain areas of the visual display (e.g., Alder, Ford, Causer, & Williams, 2014; Goulet et al., 1989; Milazzo, Farrow, Ruffault, & Fournier, 2015). Several researchers have used gaze behaviour data to infer the location from which information is extracted to facilitate anticipation (e.g., Savelsbergh, Williams, van der Kamp, & Ward, 2002; Singer, Cauraugh, Chen, Steinberg, & Frehlich, 1996; Williams, Ward, Knowles, & Smeeton, 2002). Williams et al. (2002) demonstrated that, in an anticipation task, skilled tennis players spent more time fixating more proximal areas of the opponent s body such as the trunk-hip and headshoulder region, whereas less-skilled players spent more time fixating more distal regions such as the racket. Similar findings, supportive of occlusion-based research, have been 17

18 observed in other sports (e.g., cricket [McRobert et al., 2009]; football [Savelsbergh et al., 2002]). However, the extraction of important information cannot be ascertained through the collection of gaze data alone (Hagemann et al., 2010; Mann & Savelsbergh, 2015). For example, although performers may fixate foveal vision on one area of the opponent s body or equipment, they may be simultaneously picking up information from other aspects of the opponent s movements through peripheral vision (e.g., Ripoll, Kerlirzin, Stein, & Reine, 1995; Williams & Elliott, 1999). To determine the important sources of information underpinning expert anticipation, the collection of multiple process-tracing measures or a combination of approaches is preferred (Hagemann et al., 2010; Williams & Davids, 1998). Pattern perception When performing in dynamic, time-constrained environments, being able to perceive developing sequences of plays is thought to be an important skill to have, the premise for this being that it may facilitate predictions of what is about to happen (Cañal- Bruland & Williams, 2010; Gorman et al., 2012). Researchers interested in pattern perception in sport have generally employed one of two paradigms, namely, the pattern recall (as outlined earlier) or pattern recognition approach. Originally employed in chess (e.g., Goldin, 1978, 1979; Saariluoma, 1984), the pattern recognition approach entails presenting participants of varying skill levels with a series of patterns, followed by presentation of another series of patterns, some of which have been previously viewed and some of which are novel. During the latter viewing, participants are required to indicate whether they recognise the patterns or not. Akin to the findings of pattern recall studies, experts have demonstrated a performance advantage in recognising patterns in structured action sequences (Goldin, 1978; North, Williams, Hodges, Ward, & Ericsson, 2011; Williams & Davids, 1995). Through manipulations of the information presented in 18

19 the visual display (Gorman, Abernethy, & Farrow, 2013; Williams, Hodges, North, & Barton, 2006; Williams, North, & Hope, 2012) and collection of gaze data (Gorman, Abernethy, & Farrow, 2015; North, Ward, Hodges, Ericsson, & Williams, 2009; van Maarseveen, Oudejans, Mann, & Savelsbergh, 2016), researchers have made notable attempts to ascertain the specific information sources facilitating effective pattern perception. It appears that information located more centrally in the visual display is of greater importance to these skills than more peripheral features (Gorman et al., 2015; Williams et al., 2012) although the important features are likely to vary from task to task. Probability assignment It is thought that the more extensive knowledge base possessed by experts in comparison to less-expert performers, facilitates the assignment of accurate probabilities to potential upcoming event outcomes in domain-specific situations. In comparison to the previously discussed perceptual-cognitive skills, probability assignment has received much less research attention (Loffing & Cañal-Bruland, 2017; Williams & Ward, 2007). Alain and Proteau (1978) provided evidence that the anticipation behaviour of skilled performers is shaped by the probabilities they assign to an upcoming event. The authors originally presented racket sports players (badminton, racketball, squash, and tennis) with rallies they had physically competed in. Through the use of percentage confidence scores, participants were asked to comment on the subjective probabilities they had assigned to the shots hit by the opponent. The authors observed that there was a strong relationship between the subjective probabilities they assigned to the opponent s shot and whether they displayed anticipatory movements (movement made prior to the opponent s racketball contact) in the corresponding video footage. More recently, the expert advantage in accurately assigning subjective probabilities to potential event outcomes has been demonstrated in football (Ward & Williams, 2003). Through an alternative approach, it 19

20 has been demonstrated that providing athletes with objective probabilities associated with potential event outcomes influences behaviour and performance (e.g., Alain & Proteau, 1977; Gray, 2015; Navia, van der Kamp and Ruiz, 2013). With the exception of Ward and Williams (2003), relatively little research has been conducted to compare how expert and less-expert performers assign subjective probabilities to potential event outcomes to inform their anticipation of an upcoming event. The interaction of perceptual-cognitive skills Researchers interested in determining the importance of the various perceptualcognitive skills to expert anticipation have generally examined these skills in isolation of one another (e.g., Müller, Abernethy, Eid, McBean, & Rose, 2010; Ward & Williams, 2003; Williams et al., 2006). While this approach clearly has benefits in that it allows for systematic investigation of specific skills in controlled and reproducible conditions (Williams et al., 2011), the dynamic, complex nature of many sports mean that there are likely to exist very few situations in which performers rely solely on one of the aforementioned perceptual-cognitive skills to anticipate the intentions of an opponent. Additionally, it may result in overreliance on skills that performers would not use to the same extent in a real-world situation, thereby inadvertently hindering the development of our understanding of expert anticipation (Williams & Ericsson, 2005). Williams (2009) suggested that these perceptual-cognitive skills are likely to interact in a dynamic, continuous manner to facilitate expert anticipation. Roca, Ford, McRobert, and Williams (2013) took an innovative approach to demonstrating the interaction of the different perceptual-cognitive skills during anticipation. Skilled and less-skilled soccer players participated in a representative anticipation task under near and far task constraints (i.e., the ball was either in the 20

21 defensive or offensive half of the pitch). Gaze data and retrospective verbal reports of thoughts were also collected. First, the authors observed that no one skill appeared to be employed in isolation, with the verbal report data indicating some contribution of postural cue usage, pattern recognition and probability assignment regardless of the task constraints. Second, clear differences in the relative importance of the various perceptualcognitive skills were observed depending on the task constraints, particularly for skilled participants who were more accurate overall. While skilled participants spent more time fixating the opponent, teammates, and areas of free space in the far compared to the near task and more time on the player in possession in the near compared to the far task, lessskilled participants displayed no such differences. These data were generally supported by the verbal reports. Skilled participants reported more thoughts indicating the use of pattern recognition skills in the far compared to the near task, and greater levels of postural cue utilisation and probability assignment in the near task. Findings illustrate that expert performers develop several domain-specific perceptual-cognitive skills which they employ to varying extents to meet specific task demands. The role of contextual information While researchers have invested great time and effort in determining the role of postural information in providing relevant cues to facilitate anticipation, the influence of contextual information and its role as a potential source of relevant cues has received comparatively little research attention (Cañal-Bruland & Mann, 2015). This narrow focus on postural information is at odds with the dynamic conditions expert performers encounter in real-world environments, which are more likely to involve multiple sources of information (Schlӓppi-Lienhard & Hossner, 2015). Context can be defined as the circumstances which form the setting for the event, statement or idea, and in terms of which it can be fully understood (Oxford Dictionaries, n.d.). By definition, context 21

22 therefore appears to be a useful form of information for understanding, and in turn evaluating and predicting the likely outcome of an upcoming event. For example, in medical diagnosis, the provision of contextual information has been shown to increase accuracy and decrease processing time in comparison with when it is not available (e.g., McRobert et al., 2013; Verkoeijen, Rikers, Schmidt, van de Wiel, & Kooman, 2004). Given the particular importance of fast and accurate responses in sports involving extreme time constraints, the limited amount of research examining the contribution of contextual information to anticipation is surprising. Buckolz, Prapavesis, and Fairs (1988) originally suggested that athletes rely on two broad types of information to anticipate an opponent s intentions; postural and contextual information. While it is acknowledged that pertinent postural cues that facilitate anticipation are only available during the opponent s action execution, some sources of contextual information which may facilitate anticipation are available earlier (i.e., prior to the opponent commencing the action). Moreover these sources of contextual information continue to be available while the opponent executes the action. Although the designs of experimental research studies examining expert anticipation are increasingly incorporating both types of information (e.g., Farrow & Reid, 2012; Loffing, Sölter, Hagemann, & Strauss, 2016; Runswick, Roca, Williams, Bezodis, & North, 2017), traditionally, researchers have controlled the potential effects of contextual information by, for example, presenting participants with isolated actions played by numerous different players from a neutral playing position (e.g., Abernethy & Russell, 1987; Renshaw & Fairweather, 2000; Savelsbergh, van der Kamp, Williams, & Ward, 2005). Early, field-based research conducted by Abernethy and colleagues (2001) allowed inferences to be made about the type of information skilled performers use to anticipate. The researchers asked skilled and less-skilled squash players to take part in simulated 22

23 matches. Participants wore liquid crystal occlusion goggles and were informed that, at some stage in the rally, their vision would be occluded. At that point, they should attempt to complete their own shot in response to the shot they anticipated the opponent would hit. When vision was occluded within the last 200 ms leading up to the opponent striking the ball, skilled participants moved more often to the most appropriate corner of the court than less-skilled participants. This finding was interpreted as evidence that the expert advantage in picking up pertinent postural cues to anticipate effectively, as demonstrated in laboratory-based research, translates to field-based settings. Importantly, however, the anticipatory movements of skilled participants continued to be more accurate than chance even when vision was occluded over 580 ms prior to the opponent striking the ball. Given that pertinent postural cues are thought to arise closer to the moment the opponent strikes the ball, the authors suggested that the accurate anticipatory movements of these skilled performers was likely due to their extensive knowledge of event probabilities associated with contextual information such as shot preferences and sequential patterns of play. Consistent with the findings of Abernethy et al. (2001), Triolet, Benguigui, Le Runigo, & Williams (2013) demonstrated that professional tennis players frequently moved in the direction of the upcoming shot prior to the opponent striking the ball (see Figure 1.2). These anticipatory movements most often occurred when players were placed in unfavourable conditions, such as when the opponent was attacking from inside the court. Such conditions were thought to constrain the players to anticipate the intentions of the opponent in order to avoid negative consequences (i.e., not reaching the ball in time to execute a return shot). While some of these anticipatory movements occurred immediately prior to the opponent s racket-ball contact, a large portion of the movements occurred over 140 ms prior to the ball being struck. The authors inferred that the later occurring anticipatory movements could have been based on postural cues, but that the 23

24 earlier occurring responses must have been based on contextual information, given that pertinent postural cues would not be available so far in advance of the opponent striking the ball. Collectively, these findings demonstrate that the lack of research into the role of contextual information needs rectifying to provide a more complete and accurate account of expert anticipation. In particular, it appears that expert anticipation is underpinned by an ability to effectively pick up and utilise both postural cues and contextual information arising from an array of potential sources. Figure 1.2. Percentage of correct anticipatory movements as a function of response initiation time relative to the opponent s racket-ball contact in professional tennis matches (t = 0) 2. Cañal-Bruland and Mann (2015) recommended that to develop a fuller understanding of anticipatory behaviour, there is a need for further research which: a) identifies the sources of contextual information that underpin anticipation; b) examines how the different sources of contextual information and postural information influence anticipation; and c) investigates how the circumstances shape the way in which the various sources of information are used or combined. The authors suggested that doing 2 Reprinted from Journal of Sports Sciences, 31, Triolet, C., Benguigui, N., Le Runigo, C., & Williams, A. M., Quantifying the nature of anticipation in professional tennis, p. 825,

25 so would provide strong foundations for the development of an overarching theoretical framework that can explain anticipatory behaviour. While the only research which has provided any indication of the latter recommendation appears to be that conducted by Triolet et al. (2013), several researchers had begun to address the former recommendations, even before the call for more research in the area had been made. Schlӓppi-Lienhard and Hossner (2015) conducted interviews with high-skilled volleyball players with the aim of identifying the factors that contribute to anticipation and decision making in defensive situations. Using inductive content analysis, the authors identified several sources of general contextual information that skilled volleyball players use in defensive situations, e.g., the score, preceding events and player positioning. In addition to these general sources of contextual information, skilled volleyball players highlighted the use of opponent-specific contextual information. Opponent specifics differ from general contextual information in that it is individualised contextual information such as action preferences, strengths, weaknesses and characteristic tells that are specific to that opponent alone. While each of these sources of contextual information deserves research attention, this thesis will focus on establishing the importance and impact of general rather than opponent-specific contextual information. With the exception of Schlӓppi-Lienhard and Hossner (2015), most researchers interested in determining the sources of contextual information that influence anticipation have manipulated the task environment to compare anticipation performance when this information is available with when it is not. For example, Paull and Glencross (1997) investigated the impact of strategic game information on anticipation in baseball. Skilled and less-skilled baseball batters anticipated the delivery from a pitcher when the pitch count and score information was either available or not. Both groups of participants were able to use this information to decrease errors in ball location prediction and decision 25

26 time. Findings suggest that strategic game information provides relevant contextual information that batters can use to anticipate more effectively regardless of skill level. Similarly, Farrow and Reid (2012) examined the influence of game score on anticipation in tennis. Skilled and less-skilled players watched sequences of serves with a score being presented before each serve in the same way as would happen in a tennis match (the participants viewed full service games). The researchers artificially manipulated the outcome of the opponent s serve on the first point in each game, such that this serve was always hit to the same place. Skilled participants detected this pattern to anticipate effectively over 600 ms prior to the ball being struck on that point but the less-skilled participants did not pick up this contextual information. While in the study by Paull and Glencross (1997) generic knowledge about likely outcomes associated with the score is likely to have influenced anticipation (see also Cañal-Bruland, Filius, & Oudejans, 2015), Farrow and Reid s (2012) findings indicate that skilled performers, but not less-skilled, appear to develop expectations through associations between the score and the outcome over short periods of time. In addition to the score, Loffing, Stern, and Hagemann (2015) showed that skilled volleyball players are more strongly influenced by contextual information picked up from the sequence of event outcomes preceding a critical event than less-skilled counterparts. Attempts have also been made to investigate the impact of specific sources of contextual information on anticipation in representative field-based studies. For example, to determine the effect of tactical initiative on anticipation, Crognier and Féry (2007) manipulated the degree to which participants were allowed to impose their game on the opponent prior to anticipating the outcome of a passing shot. Participants wore liquid crystal occlusion goggles which were activated as the opponent was about to hit the passing shot. Upon occlusion, participants moved as if they were returning the shot. 26

27 Anticipation accuracy was highest in the high tactical initiative condition in which the participants were allowed to impose their game on the opponent over a series of shots and lowest in the low tactical initiative condition in which they had no opportunity to do so. While attempts are therefore being made to address the first recommendation put forward by Cañal-Bruland and Mann (2015), these attempts have largely been isolated attempts in which one potential source of contextual information is artificially manipulated to determine its impact on anticipation. However, in real-world situations multiple sources of contextual information will be available at any given time. Systematic investigation of the impact of the various available sources of contextual information and how they may be combined while anticipating is therefore still needed. Additionally, very little research has been conducted to investigate how contextual information is combined with postural cues during anticipation. Müller and Abernethy (2012) proposed what they referred to as a preliminary model of anticipatory skill in striking sports. In this model, they suggest that contextual information, available prior to the commencement of the opponent s movement pattern, may help prime a skilled athlete s movement response for when more confirmatory postural cues or ball flight information become available. Conversely, less-skilled athletes are thought to rely on late occurring postural cues and ball flight information to inform their judgments. Additionally, they proposed that, in striking sports at least, a progressive switching from the utilisation of contextual to postural information sources is required for effective anticipation. However, this potential interaction between the use of contextual information and postural cues in the lead-up to the critical event is yet to be sufficiently investigated. A recent exception to this is the research conducted by Loffing and Hagemann (2014). The authors presented skilled and less-skilled tennis players with point-light displays of an opponent hitting down-the-line or cross court shots. The position of the 27

28 opponent and the moment of occlusion relative to the opponent s racket-ball contact were systematically varied in an attempt to a) examine the influence of contextual information on expectancies of shot outcome and b) investigate the potential interaction between contextual (court positioning) and postural (opponent s movement pattern) information during anticipation. The authors demonstrated that contextual information picked up from an opponent s court positioning influenced expectancies during anticipation and that skilled participants were more strongly influenced by this contextual information than less-skilled participants. Specifically, the further away from the centre line the opponent was positioned, the more likely skilled participants were to anticipate the opponent to direct the shot cross-court. Moreover, the authors observed that the expectancies of skilled participants were most strongly influenced by contextual information early in the development of the opponent s movement pattern. Loffing and Hagemann interpreted this as evidence that skilled performers more extensive domain-specific knowledge bases allowed them to adjust their expectancies of potential event outcomes based on the contextual information picked up from the opponent s court positioning. The less extensive knowledge bases of less-skilled participants, on the other hand, did not allow them to integrate this information into their judgment of the potential upcoming event to the same extent as the skilled participants. McRobert, Ward, Eccles, and Williams (2011) investigated whether the perceptualcognitive processes underpinning anticipation differed depending on the availability of contextual information. Skilled and less-skilled cricket batters were required to anticipate the intentions of bowlers in two conditions in which the level of contextual information available was varied. In the low-context condition, participants viewed footage of bowls from six different bowlers presented in a random order, whereas in a high-context condition participants viewed four different bowlers, with all bowls from each bowler 28

29 presented in blocks of six, the rationale being that contextual information could be picked up from the action tendencies of the opponent in the high-context condition. The authors observed that both groups performed more effectively in the high- compared to the lowcontext condition. The skilled participants but not the less-skilled, also adjusted their gaze strategy to extract information from relevant cues more efficiently in the high context condition. Moreover, consistent with Long Term Working Memory (LTWM) theory (Ericsson & Kintsch, 1995), the skilled participants made more evaluation, prediction and deep planning statements than less-skilled participants suggesting that the employment of more elaborate domain-specific memory representations facilitated this adjustment in gaze strategy. Finally, the authors suggested that these representations allowed the skilled participants to use the preceding context to rapidly encode and retrieve task-relevant information during anticipation. Cognitive processes underpinning expert anticipation and decision making Ericsson and Kintsch (1995) proposed LTWM theory to account for some processes displayed during expert performance not accounted for by Chase and Simon s theory of expertise (1973). Chase and Simon originally suggested that expertise was characterised by immediate access to relevant knowledge. In recognising structured configurations of chess pieces, these configurations acted as cues to provide access to the best possible moves, which had been stored in Long Term Memory (LTM) from previous experience with that configuration of pieces. The most notable limitation of their theory was the suggestion that expert performers relied on STM for temporary storage of chunks of information. Subsequent to the seminal research of Chase and Simon (1973), expert performers were shown to not only be able to maintain high levels of recall of domainspecific information despite interpolation in a recall task (Charness, 1976), but that they could rapidly encode large amounts of information (Frey & Adesman, 1976), and rapidly 29

30 retrieve encoded information from LTM (Anderson, 1990). Ericsson and Smith (1991) concluded that the mechanism put forward by Chase and Simon (1973) was overly restrictive as it proposed that the processing demands of working memory were constrained to work within the limits of STM. In light of these developments, Ericsson and Kintsch (1995) proposed that expert performers develop skills which allow them to circumvent or extend the limited capacity of working memory based on STM. Through experience and practice (Ericsson et al., 1993), expert performers acquire retrieval structures that facilitate encoding of domainspecific information with retrieval cues, which are related to prior knowledge and facilitate the retrieval of large amounts of relevant information from LTM. This process becomes more efficient with deliberate practice in the domain (Ericsson et al., 1993), in that the time taken to encode and retrieve domain-specific information reduces to a point comparable with that which would be expected if the information was being retrieved from STM. In addition to proposing an explanation of how expert performers circumvent the limitations of STM in pattern recall or other memory tasks, LTWM theory (Ericsson & Kintsch, 1995) proposes an insight into the cognitive mechanisms underlying expert performance in complex tasks requiring efficient anticipation and decision making skills. Specifically, an important feature of the retrieval structure is that upon activation during encoding, it facilitates prediction of future retrieval demands, and the generation and evaluation of multiple relevant options, which ultimately enhances decision quality (Ericsson, Patel, & Kintsch, 2000). Indicative of their more elaborate domain-specific memory representations, in representative tasks, expert performers generally report engagement in more evaluation and forward planning than less-expert counterparts (Ericsson et al., 2000), rather than the mere monitoring of immediately available information. Through the use of think-aloud 30

31 verbal report protocols (Ericsson & Simon, 1993), researchers have demonstrated that the superior performance of experts is underpinned by these processes in sport (e.g., McRobert et al., 2011; Roca, Ford, McRobert, & Williams, 2011), law enforcement (Ward et al., 2011), and medical diagnosis (McRobert et al., 2013). More recently, option generation paradigms have been employed during anticipation and decision making tasks to demonstrate that, expert performance is associated with the generation of not only the best or most likely option but also potential relevant alternatives (Belling, Suss, & Ward, 2015a, Suss & Ward, 2012; Ward, Ericsson, & Williams, 2013). Collectively, these findings are consistent with the notion that the retrieval structures developed by expert performers do not prescribe a set response but rather they allow access to potential relevant alternatives, facilitating higher quality decisions (Ericsson et al., 2000). Alternative models of expert decision making propose that increased levels of expertise are associated with lower levels of cognitive evaluation. As part of his seminal research on expertise in chess, de Groot (1965) highlighted that the superior performance of expert chess players was often characterised by quick recognition of the next best move, rather than generation and evaluation of potential alternative moves. More recently, Klein and colleagues proposed the Recognition Primed Decision (RPD) model (e.g., Klein, 1993; Klein, Calderwood, & Clinton-Cirocco, 1986). The RPD model suggests that, through repeated exposure to a particular situation, expert performers develop associations between the information available within the environment and the optimal response. When similar information is encountered in subsequent performance, the information is recognised and matched to an appropriate response. Alternative responses are only generated if, having mentally simulated that response, it is deemed inappropriate. Based on similar principles as the RPD model, Johnson and Raab (2003) later proposed the Take The First (TTF) heuristic. In contrast to the evaluation and 31

32 forward planning LTWM theory (Ericsson & Kintsch, 1995) associates with expert performance, the TTF heuristic model recommends that expert performers should take the first option that comes to mind, as this will be the highest quality option generated. Raab and colleagues (e.g., Johnson & Raab, 2003; Raab & Johnson, 2007; Laborde & Raab, 2013) have provided support for the TTF heuristic model in a series of experiments on expert decision making in handball. At this point however, our knowledge of the cognitive processes underlying the use of contextual information during expert anticipation, as well as how this information is combined with postural information, is limited (for exceptions see Loffing & Hagemann, 2014; McRobert et al., 2011; Milazzo et al., 2015). Aims of the Thesis The overall aim of the current thesis is to examine the role of contextual information in expert anticipation, with tennis being used as a vehicle to explore the issue. To achieve this aim, comparisons will be made between skilled and less-skilled tennis players over the course of five studies. The extent to which contextual information can be used to anticipate the intentions of an opponent relative to when pertinent postural cues are also available will be investigated, the specific sources of contextual information contributing to anticipation explored and the mechanisms underpinning anticipation under different informational constraints examined. Thus far, researchers have only been able to infer, based on the timing of anticipatory movements, that expert performers can anticipate effectively based on contextual information picked up in advance of pertinent postural cues becoming available (Abernethy et al., 2001; Triolet et al., 2013) and little is known about how they may do so. In Chapter 2 (Studies 1 and 2), an attempt is made to determine the extent to 32

33 which tennis players can anticipate the outcome of an opponent s shot when presented with contextual information alone, compared with when pertinent postural cues are also available for processing, and to investigate the processes underpinning skilled anticipation when constrained to anticipate based on contextual information alone. In Study 1, skilled and less-skilled tennis players will be presented with open-play rallies from real tennis matches in two conditions; a video condition in which all information that would normally be available to players is available (i.e., both contextual and postural information), and an animated condition which has been edited to omit the bodies and rackets of the players such that participants are constrained to anticipate based on contextual information alone (e.g., player positioning, shot sequencing). It is expected that both contextual and postural information will contribute to accurate anticipation and that skilled participants will be able to use both types of information more effectively than less-skilled counterparts. In Study 2, gaze data and verbal reports of thoughts will be collected to examine the perceptual-cognitive processes underpinning anticipation when constrained to anticipate based on contextual information alone. It is expected that skilled participants will display different gaze behaviour (Mann et al., 2007) to less-skilled participants as well as thought processes supportive of Long Term Working Memory theory (Ericsson & Kintsch, 1995) as an explanation of expert anticipation. In light of the first recommendation made by Cañal-Bruland and Mann (2015), in Chapter 3 (Studies 3 and 4), an attempt is made to identify the specific sources of contextual information which facilitate anticipation. Skilled and less-skilled tennis players will be presented with animated footage from which they will be constrained to anticipate based on contextual information alone. The test stimuli viewed by participants will be manipulated to examine the extent to which various sources of contextual information (shot sequencing, ball and player motion and positioning) facilitate 33

34 anticipation. In Study 3, participants will anticipate the outcome of an opponent s shot in three conditions in which the number of shots in the sequence preceding the final occluded shot is systematically varied. It is expected that participants will be more accurate when the preceding shot sequence is presented than when it is not and that skilled participants will use this information more effectively than less-skilled counterparts. In Study 4, participants will be presented with sequences of a fixed length which depict only the ball, the players, or both the ball and the players, in dynamic or still form. Overall, decrements or increases in performance levels when a particular source of information is omitted or made available in comparison to a control condition will provide an indication of the extent to which that source of contextual information facilitates anticipation. While several researchers have investigated the cognitive processes underpinning expert anticipation, how expert performers generate options prior to selecting the option considered most likely to occur has received surprising little research attention, particularly considering the importance of this process in the natural ecology (Klein, 1993). In Chapter 4 (Study 5), the option generation strategies of skilled and less-skilled tennis players when anticipating the outcome of an opponent s shot will be investigated. As in Study 1, participants will view animations of rallies in which they are constrained to anticipate based on contextual information alone or videos in which pertinent postural cues are also available for processing. It is expected that the option generation strategies employed by expert performers will be consistent with LTWM theory (Ericsson & Kintsch, 1995), in that the generation of relevant alternative options in addition to the most likely outcome will result in more accurate anticipation. The use of an option generation paradigm in this study will further provide an initial insight into how expert performers integrate contextual and postural information to facilitate anticipation (Cañal- Bruland & Mann, 2015). 34

35 Finally, in Chapter 5, an epilogue is presented which is used to illustrate and bring together the main findings of the thesis. It will further be used to demonstrate how the thesis has contributed to the development of knowledge, both theoretically and from an applied perspective, as well as to identify limitations of the research programme. Moreover, future research directions will be outlined, with a clear focus on how researchers can extend this and other emerging research to develop a fuller understanding of this relatively nascent research area. Methods A range of methods will be employed in the forthcoming experimental chapters to address the specific aims of the thesis, some of which are commonly employed by researchers and some more novel. To investigate the impact and importance of contextual information during skilled anticipation, throughout the thesis, participants will be presented with novel test stimuli. Player movement and ball trajectory data from real tennis matches will be used to create animations that simulate rallies played between two players while omitting postural information and the rackets of the players from the visual display. The presentation of test stimuli that omits postural information from the display will thereby constrain participants to anticipate based on contextual information alone. A similar approach has been taken to investigate how biological motion information can be used to anticipate an opponent s intentions (e.g., Abernethy et al., 2008; Huys et al., 2009; Ward, Williams, & Bennett, 2002). In Studies 2 and 5, the processes and mechanisms underpinning skilled anticipation will be examined. Gaze behaviour and verbal report data will be collected in Study 2 to examine the perceptual-cognitive processes underpinning skilled anticipation, specifically, when constrained to anticipate based on contextual information alone. While 35

36 gaze behaviour data will provide an indication of the information source being fixated on, verbal reports will provide an indication of how participants cognitively process this information prior to making a response. These methods have frequently been employed by researchers interested in further understanding the perceptual-cognitive processes underpinning skilled anticipation (e.g., McRobert et al., 2009; North et al., 2011; Roca et al., 2011). Conversely, the option generation paradigm, as employed in Study 5 has been employed much less frequently. Option generation tasks allow researchers to examine the cognitive strategies that support skilled anticipation and decision making, however the majority of researchers employing this approach have examined decision making rather than anticipation (for exceptions see Belling et al., 2015a; Belling, Suss, & Ward, 2015b). In option generation tasks, rather than being required to choose between a set number of options, participants indicate the options they generated and considered prior to anticipating the action outcome, such as would be necessary in the natural ecology (Klein, 1993). This approach therefore facilitates further examination of the cognitive strategies that support anticipation in real-world situations. In line with Ericsson and Smith s (1991) Expert Performance Approach, researchers now commonly attempt to capture expert performance in laboratory-based settings using life-size video simulations (e.g., McRobert et al., 2009; Roca et al., 2011; Williams et al., 2002) and movement-based responses (e.g., McRobert et al., 2009; Roca et al., 2013). The use of large screens onto which test stimuli are projected rather than smaller screen sizes is based on the assumption that this creates a more realistic environment for the participant (Williams & Davids, 1998). Similarly, in comparison to pen and paper responses that were used in earlier research (e.g., Abernethy & Russell, 1987; Buckolz et al., 1988; Jones & Miles, 1978), requiring participants to move while responding to test stimuli is proposed to increase the fidelity of the task (Roca et al., 36

37 2013). Where feasible, an attempt is made to maintain a realistic task environment throughout the thesis, while nevertheless maintaining acceptable levels of internal validity and experimental control. 37

38 Chapter 2 Contextual information and perceptual-cognitive expertise in a dynamic, temporally-constrained task 38

39 Abstract Skilled performers extract and process postural information from an opponent during anticipation more effectively than their less-skilled counterparts. In contrast, the role and importance of contextual information in anticipation has received only minimal attention. We evaluate the importance of contextual information in anticipation and examine the underlying perceptual-cognitive processes. We present skilled and lessskilled tennis players with video or animated footage of the same rallies. In the animated condition, sequences were created using player movement and ball trajectory data, and postural information from the players was removed, constraining participants to anticipate based on contextual information alone. Participants anticipated ball bounce location of the opponent s final occluded shot. The two groups were more accurate than chance in both display conditions with skilled being more accurate than less-skilled participants (Study 1). When anticipating based on contextual information alone, skilled participants employed different gaze behaviours to less-skilled counterparts and provided verbal reports of thoughts which were indicative of more thorough evaluation of contextual information (Study 2). Findings highlight the importance of both postural and contextual information in anticipation and indicate that perceptual-cognitive expertise is underpinned by processes that facilitate more effective processing of contextual information, in the absence of pertinent postural cues. 39

40 Introduction Researchers interested in anticipation in real-world tasks often use tennis as a vehicle to explore the underlying mechanisms due to the extreme time constraints involved. Ball speeds can reach over 200 km h -1 (Gillet, Leroy, Thouvarecq, & Stein, 2009). The time it takes for the ball to travel from one player to the other can be as little as 500 ms, meaning it is sometimes impossible for players to respond quickly enough to return the opponent s shot based on ball flight information alone (Abernethy & Wollstein, 1989). The extreme time constraints evident in fast ball sports (Gray, 2002a), coupled with the fact that an opponent can potentially disguise his/her intentions or present deceptive information (Jackson, Warren, & Abernethy, 2006; Kunde, Skirde, & Weigelt, 2011; Rowe, Horswill, Kronvall-Parkinson, Poulter, & McKenna, 2009), suggest that the contextual information which is available prior to pertinent postural cues becoming available could be both valuable and necessary when making anticipation judgments. The majority of researchers focusing on anticipation, particularly in racket-sports, have focused on how performers use advance postural cues to inform their judgments (for reviews see Crognier & Féry, 2007; Mann et al., 2007). However, recent calls have been made for more research investigating the role of contextual information in anticipation (Cañal-Bruland & Mann, 2015, Triolet et al., 2013). It is well established that skilled performers are better able to extract and process postural information from an opponent to more accurately (e.g., Abernethy & Russell, 1987; Jones & Miles, 1978) and more quickly (e.g., Ward et al., 2002; Williams et al., 2002) anticipate the outcome of an upcoming event than their less-skilled counterparts. To rigorously examine how skilled performers more effectively utilise postural information, researchers have traditionally employed quasi-experimental cross-sectional designs (comparing skilled and less-skilled performers) in which participants anticipate the outcome of an event based solely on the 40

41 postural information emanating from an opponent (e.g., Goulet et al., 1989; Jackson & Mogan, 2007). In such experimental set-ups, contextual information built up prior to the event is usually omitted, such that the opponent s positioning is limited to a single location of the playing area throughout trials. However, published research suggests that skilled performers can effectively anticipate the outcome of an upcoming event based on contextual information picked up in advance of pertinent postural cues becoming available (Abernethy et al., 2001; Triolet et al., 2013). Moreover, contextual information can be used by skilled performers to anticipate more effectively than when it is not available, (e.g., Crognier & Féry, 2005, Farrow & Reid, 2012). Several researchers (e.g., McRobert et al., 2009; Roca et al., 2011) have explained the processes underpinning skilled anticipation using Long-Term Working Memory (LTWM) theory (Ericsson & Kintsch, 1995). Nevertheless, both our understanding of the extent to which performers can use contextual information to anticipate effectively in advance of pertinent postural cues becoming available, and the perceptual-cognitive processes underpinning such judgments is limited. Triolet et al. (2013) quantified anticipation behaviour in professional tennis matches. A key finding of this research was that while some anticipation behaviour was found to occur around the time of the opponent s racket-ball contact, a large portion of effective anticipation behaviour occurred over 140 ms prior to the opponent striking the ball. Moreover, a large portion of this early anticipation behaviour occurred when the receiving player was under unfavourable conditions, that is, under extreme time constraints. If a visual-motor delay of approximately 200 ms (Hick, 1952) is assumed, players must have been anticipating based on information arising at least 340 ms prior to the opponent s racket-ball contact. Triolet and colleagues concluded that these highly skilled performers were using contextual information to anticipate in advance of pertinent 41

42 postural cues becoming available. These findings indicate that contextual information may play an important role in skilled anticipation, particularly in fast ball sports such as tennis, in which players often perform under extreme time constraints, and it is not viable to wait for pertinent postural cues to become available. In a related study Abernethy et al. (2001) demonstrated that skilled squash players are able to anticipate shot direction and depth at levels significantly greater than chance over 580 ms before the opponent s racket-ball contact. They concluded that the skilled squash players performance advantage was due to their superior attunement to information picked up from within their opponent s pattern of play (e.g., accurate knowledge of event probabilities, shot sequencing information). Similarly, Loffing and Hagemann (2014) demonstrated that the positioning of the opponent relative to court markings provides contextual information about the direction of the opponent s shot, and that skilled players rely more on this positional information in advance of pertinent postural cues becoming available. This body of research indicates that skilled performers can use early advance information to anticipate effectively (Abernethy et al., 2001; Triolet et al., 2013) and that they do so when placed under extreme time constraints such as when the opponent is attacking or when the distance between the two players is reduced (Triolet et al., 2013). It is not yet clear though, whether this advance information is picked up from very early occurring postural cues or from contextual information, because postural information has always been readily available to participants. While postural information from the opponent will always be available in competition as the opponent s body is always visible, the moment at which this information can reliably inform skilled performers anticipation judgments varies across sports and situations (Abernethy et al., 2001). It is 42

43 therefore difficult to infer the type of information performers are using based solely on how far in advance of the event they are acting on an anticipation judgment. Context can be defined as the circumstances that form the setting for an event, statement, or idea, and in terms of which it can be fully understood (Oxford Dictionaries, n.d.). Contextual information can come in many different forms, such as knowledge of an opponent s strengths/weaknesses, players positioning relative to one another and climatic conditions (Buckolz et al., 1988). The positive effect of providing contextual information to participants in addition to postural information has been demonstrated in several sports (for a review, see Cañal-Bruland & Mann, 2015). Researchers have demonstrated higher levels of anticipation performance when participants are provided with contextual information in the form of scores and number of balls and strikes in baseball (Paull & Glencross, 1997), increased exposure to the bowler in cricket (McRobert et al., 2011) and knowledge of the opponent s action preferences in soccer (Navia et al., 2013). In tennis, Crognier and Féry (2005) demonstrated that players anticipated more accurately in a high context, which they termed high tactical initiative, condition in which they were allowed to impose their game on the opponent over a series of shots, compared to when they were not allowed to do so and no shots were played prior to the opponent s occluded passing shot. Farrow and Reid (2012) demonstrated that skilled tennis players can use knowledge of the opponent s serving tendencies to increase the speed of their anticipation judgments. Conversely, researchers have found that presenting performers with contextual information can also yield a decrease in performance. In a simulated baseball batting task, Gray (2002a, 2002b) demonstrated that while contextual information picked up from the preceding sequence of pitches and pitch count influences batting accuracy, this information negatively affected performance when incongruence existed between the 43

44 actual and expected pitch. Mann, Schaefers, and Cañal-Bruland (2014) and Loffing et al. (2015) have produced supportive findings of such an effect in handball and volleyball anticipation judgment tasks. Collectively, these findings suggest that while contextual information is often beneficial to performance, situations exist when attending to contextual information can be misleading and the use of reliable postural information may be more favourable. However, under extreme time constraints, it is not always feasible for performers to wait for such information to act. As opposed to manipulating contextual information when postural information is readily available, we provide a novel contribution by comparing anticipation response accuracy either in the presence or absence of postural information at the same moment so that conclusions can be drawn about the extent to which performers can use contextual information to anticipate effectively. We consider contextual information to constitute the circumstances which form the setting for the event; circumstances which in advance, or in the absence of pertinent postural cues would necessarily be relied upon to make anticipation judgments. Therefore, contextual information in our anticipation task refers to the sequential relative movements of the players and the ball flight in the lead up to the critical event (the to-be-anticipated shot played by the opponent), and the players resultant positioning at the moment of the event. In contrast, postural information refers to the bodily movements of the players and their resultant racket movements. In Study 1, we investigate the extent to which tennis players can anticipate based on contextual information alone compared with performance based on this information as well as postural information. Moreover, because skilled performers have previously been shown to anticipate more effectively than their less-skilled counterparts in advance of pertinent postural cues becoming available (Abernethy et al., 2001), in Study 2 we aim to determine 44

45 how they do so, by tracing the perceptual-cognitive processes underlying judgments when constrained to anticipate based on contextual information alone. Process-tracing measures such as the recording of gaze behaviour, collection of verbal reports of thoughts, and manipulations to representative tasks are often used to identify the mechanisms underpinning skilled performance (e.g., Afonso, Garganta, McRobert, Williams, & Mesquita, 2012; Catteeuw, Gilis, Wagemans, & Helsen, 2010; Ward et al., 2013). A comprehensive understanding of the perceptual-cognitive processes underlying anticipation provides an insight into how skilled performers process information more effectively than their less-skilled counterparts. This information can then be used to guide practice organisation for training and development purposes (Williams et al., 2011). Ericsson and Kintsch s (1995) Long Term Working Memory (LTWM) theory states that as a result of extended experience and practice, skilled performers have developed complex domain-specific memory representations. When they encounter information during practice or competition, they can accurately encode and store this information in LTM. The information is associated with a retrieval cue in STM, meaning that when similar situations are subsequently encountered, skilled performers can rapidly access the associated information stored in LTM through these retrieval cues. Skilled performers can therefore access information about previous outcomes in similar situations to guide their anticipation judgments. During practice skilled performers are exposed to a diversity of events and situations (e.g., the opponent in a tennis practice match hitting an attacking forehand from inside the court), and are provided with the opportunity to respond to these events (e.g., by anticipating the depth and direction of the opponent s shot and attempting to return the shot). This gives the skilled performer the opportunity to encode information from these situations, associate this encoded information with a retrieval cue, and store 45

46 the information in LTM. When skilled performers encounter a similar situation in subsequent performance, the presented information is associated with a retrieval cue, allowing for the rapid retrieval of relevant information about that situation and potentially the previous outcomes of such a situation from LTM. Less-skilled performers, on the other hand, have not developed such elaborate memory representations due to their relative lack of experience. Several researchers have provided support for LTWM theory (Ericsson & Kintsch, 1995) in domain-specific anticipation tasks (e.g., North et al., 2011; Roca et al., 2011; Ward et al., 2013), citing skilled performers more elaborate domainspecific memory representations as the reason for differences in performance between skill groups. Verbal reports of thoughts recorded during (concurrently) or immediately following (retrospectively) an anticipation task provide an insight into performers task-specific knowledge and cognitive processes when carrying out a task (Anderson, 1987). Ericsson and Simon s (1993) protocol analysis method has frequently been used to categorise the cognitive processes underlying skilled performance. In representative domain-specific anticipation tasks, skilled performers have been shown to verbalise more thoughts relating to evaluation of the event and prediction of potential future outcomes than their lessskilled counterparts (e.g., McRobert et al., 2011; North et al., 2011; Roca et al., 2011). The higher number of evaluation and prediction statements made by skilled performers when reporting their thoughts have been interpreted as being supportive of LTWM theory (Ericsson & Kintsch, 1995). It has been suggested that skilled performers more elaborate memory representations in domain-specific situations direct their gaze to the more pertinent information present in the environment (Williams & Elliott, 1999). Increasingly, researchers have collected data for perceptual and cognitive process measures in the same 46

47 study so as to establish much more fully the processes contributing to performance differences (Afonso & Mesquita, 2013; McRobert et al., 2011). While the visual search strategy employed is dependent on the task-constraints (Roca et al., 2013; Williams, Janelle, & Davids, 2004), skilled performers have usually been shown to employ different strategies to those of their less-skilled counterparts (for a review, see Mann et al., 2007). In a representative cricket task, McRobert et al. (2011) varied the number of times in a row participants viewed a particular bowler to compare the perceptual-cognitive processes employed by batters in low- and high-context conditions. Verbal report data revealed that the batters made more evaluation and deep planning statements when contextual information was available about the bowler being viewed and skilled batters made more evaluations, predictions and deep planning statements in both conditions, which the authors suggested contributed to their superior performance on the task. Gaze data indicated that skilled batters adapted their visual search strategies in the high-context condition to more efficiently extract pertinent information from the visual display than in the low-context condition. These data were interpreted to mean that in high-context conditions, skilled batters are able to use contextual information (in this case picked up from an opponent s prior actions), to rapidly encode and retrieve task-relevant information stored in LTM. However, thus far, no researchers have investigated whether skilled performers use similar processes when presented with contextual information alone, as would be the case when performers are under extreme time constraints and pertinent postural cues picked up from the opponent are therefore not available. Although the aforementioned studies provide us with valuable information about perceptual-cognitive expertise, to our knowledge postural information and/or incoming ball flight information has always been readily available to the participant when carrying out the task, making it possible for participants to base their judgments solely on the 47

48 availability of such information. This approach may be somewhat misleading when we consider the multidimensional nature of anticipation (i.e., the potential importance and contributions of various sources of information and constraints placed on performers in the anticipation process, Müller & Abernethy, 2012; Roca et al., 2013). Triolet et al. (2013) reported that skilled performers often act in advance of pertinent postural cues becoming available due to the extreme time constraints involved. As such, it is important to determine how skilled performers process the information that would necessarily be relied upon when pertinent postural cues are not available. In the current paper, we first investigate the extent to which skilled and less-skilled tennis players can use contextual information to anticipate. Second, we examine the perceptual-cognitive processes underlying anticipation judgments based on contextual information alone. Study 1 We used a laboratory-based simulation of an anticipation task to compare the performance of skilled and less-skilled tennis players when viewing test stimuli which either omitted postural information, so as to present only the sequential relative movements of the players and the ball flight in the lead up to the opponent s shot (animated condition), or which presented this contextual information as well as postural information (video condition). First, we examined whether the presentation of contextual information alone allows for effective anticipation or if postural information is necessary to anticipate effectively. Based on the findings of Abernethy et al. (2001) and Triolet et al. (2013) we hypothesised that all tennis players would be able to anticipate at levels significantly greater than chance in both video and animated conditions. Second, we determined the extent to which providing postural information in addition to contextual information affects anticipation accuracy. We hypothesised that participants would be more accurate in the video condition than in the animated condition due to the availability 48

49 of pertinent postural cues (Müller & Abernethy, 2012). We further hypothesised that the increase in accuracy from animated to video condition would be greater in the skilled than the less-skilled group due to the skilled group s greater ability to pick up and utilise pertinent postural cues from an opponent (Mann et al., 2007). Third, we investigated whether skilled participants would be more accurate than their less-skilled counterparts when presented with contextual information alone, as well as when presented with both contextual and postural information. We hypothesised that skilled participants would be more accurate than less-skilled participants in both the video condition in which postural information was readily available (Mann et al., 2007) and in the animated condition in which contextual information would be used to inform judgments (Abernethy et al., 2001; Loffing & Hagemann, 2014). Methods Participants Altogether, 16 skilled (Mage = 24.0 years, SD = 5.6) and 20 less-skilled (Mage = 24.1 years, SD = 4.7) male tennis players participated. Skilled participants had a mean of 17.8 (SD = 5.5) years of tennis playing experience, 12 of whom held an Association of Tennis Professionals (ATP) singles ranking (mean career high ranking of 671 [SD = 418] in the world). Less-skilled participants had a mean of 7.0 (SD = 4.8) years of tennis playing experience. Skilled and less-skilled participants had played a mean of 13.3 (SD = 3.6) and 1.4 (SD =.6) hours tennis per week throughout their career respectively. Skilled participants had competed in 19.1 (SD = 7.3) tournaments per year, whereas less-skilled participants had not competed in competitive tennis tournaments outside of internal staff box leagues or schools tennis. Two participants in each group were left-handed with respect to the hand they normally use to play tennis, whereas all other participants were 49

50 right-handed. All participants reported normal or corrected vision and those with corrected vision wore contact lenses or glasses. Participants gave their written informed consent prior to taking part and were informed that they could withdraw at any time without penalty. The work was carried out according to the ethical guidelines of the lead university. Test Stimuli Footage of professional men s tennis matches was collected at the AEGON Championships (2013) at The Queen s Club, London from a height of 1.9 m above the ground, 6.4 m behind the centre of the court s baseline, using a 30 Hz wide angle HD video camera (Contour Roam, Contour Inc., Seattle, USA). The camera was positioned such that the two players movements and ball flight were captured with the opponent whose shot the participant was anticipating in the test stimuli always being the player positioned on the far side of the court. A total of 11 matches were recorded from various rounds of the tournament, which provided footage of 15 different players, all of whom were right-handed. Test stimuli were created in two display conditions, namely video and animated (see Figure 2.1). Animated trials were created using player movement and ball trajectory information of the same points used in the video condition. A 10-camera system positioned around the court during matches tracked the ball s trajectory and players movements throughout points (Hawk-Eye Innovations Ltd., Basingstoke, England). Ball flight information was in the form of trajectory data (x, y, z coordinates as a function of time) while player movement data were in the form of x, y coordinates. The data from selected points were input into the LTA rendering engine (Julien Pansiot, London, UK, 2013) through which animated videos of the data were generated for use with VLC media 50

51 player (VideoLAN, Paris, France). The principal difference between the two display conditions was that postural information of the players was not visible in the animated condition. In place of seeing the players and their rackets, participants saw a blue and a red cylinder representing each player and the ball as a yellow dot, while rackets were not visible. In both display conditions trials were occluded at the opponent s racket-ball contact using Pinnacle Studio 15 editing software (Pinnacle, Ottawa, Canada). Figure 2.1. Video (top) and animated (bottom) display conditions. 51

52 Experimental trials were made up of footage from 90 rallies, with the final occluded shot played by the opponent to one of four areas on the court. Those shots the opponent hit on which the first bounce landed before or after the service line were classified as short and deep respectively, while those shots which landed to the right or left of the centre line were classified as such. The number of shots played to the four areas of the court on the final occluded shot was: deep-left (32); deep-right (32); short-left (13); and short-right (13). The order of the bounce location of the opponent s final occluded shot was randomised across trials and conditions. The number of shots played prior to the opponent s occluded shot in trials was between 2 and 12. Trials were between 5 and 18 seconds long, beginning 3 seconds prior to racket-ball contact of the serve at the beginning of the point. An inter-trial interval of 6 seconds was employed. Three Lawn Tennis Association (LTA) tennis coaches, all of whom were coaching national and/or international players on a full-time basis, recommended presenting situations in which the receiving player (on the near side of the screen) was placed under extreme time constraints as test stimuli. The rationale for doing so was that in a match situation such extreme time constraints are likely to force the player to anticipate the outcome of the opponent s shot early, in order to return it successfully, as it may not be feasible to wait for pertinent postural cues or ball flight information to become available. In support of this practical recommendation, Triolet et al. (2013) reported that in professional men s tennis matches, anticipation behaviour is more prevalent when the balance of power is unfavourable for the receiving player, and when the spatiotemporal constraints are therefore more extreme. The selected situations in which trials were occluded at the opponent s racket-ball contact were made up of attacking forehand and backhand drives, passing shots, volleys, and drop-shots, all of which meant that the receiving player was under extreme time constraints, the balance of power was 52

53 unfavourable for him, or both. A total of 150 points (150 videos and the corresponding 150 data sets to be used to render animations) were originally selected, with points being omitted if the footage was of low quality, due to poor weather and lighting, or if the data used to render the animations contained errors, such as gaps in ball flight information. We thereby reduced the number of trials from 150 to 90, which were viewed in both video and animated display conditions (90 videos and 90 animated trials). Materials, Apparatus, and Set-Up Footage was projected on to a m white projection screen (AV Stumpfl, Wallern, Austria) using a NEC PE401H HD projector (NEC, Tokyo, Japan). Participants began each trial standing 5 m from the screen holding a racket in their hands as if ready to play a point. Testing was conducted in a large enough room for participants to be able to move around and swing the racket freely. The screen further allowed for recreation of a realistic vertical visual angle of subtended by the opposing player. A similar visual angle has been used in other studies of this nature, (e.g., Loffing, Wilkes, & Hagemann, 2011). Procedure Participants viewed 16 familiarisation and 180 experimental trials, half of which were shown in the video condition and half in the animated display condition. The rallies used in familiarisation trials were different to those used in experimental trials. Participants viewed trials in blocks of 30 with a one minute break between each of the blocks except between the third and fourth blocks for which they had a three minute break (merely to have a longer break half-way through the protocol). The order of display conditions was counterbalanced across participants: half the participants in each group viewed a block of 30 animated trials followed by five further blocks of 30 clips that 53

54 alternated between video and animated trials. The remaining participants viewed a block of 30 normal trials followed by five further blocks of 30 trials that alternated between animated and video. Prior to commencing the protocol, participants were informed that each of the occluded shots played by the opponent landed in the legal playing area, in one of the four sections of the court (deep-left, deep-right, short-left, short-right). During trials participants were allowed to move freely. Upon occlusion at the opponent s racket-ball contact, the screen would go black and participants verbally indicated depth and direction of the ball bounce location, while swinging the racket and moving as if to return the shot. The verbal response was recorded. Participants did not receive feedback after any trials. Data Analysis Response accuracy was reported as the percentage of correct anticipation judgments relative to the actual final ball bounce location of the occluded shots. To determine whether participants could anticipate more effectively than chance, one-sample t-tests were carried out to compare depth, direction, and combined response accuracy of the two groups with response accuracy that would be expected due to chance. To examine the extent to which skilled and less-skilled participants can use contextual information alone and both contextual and postural information to inform their judgments, a 2 2 (Display [video, animated] Group [skilled, less-skilled]) multi-factorial MANOVA with repeated measures was conducted for response accuracy, with the percentage of correct depth, direction, and combined judgments serving as the dependent variables. Wilks Lambda values are reported for the multivariate output while the Greenhouse-Geisser correction was applied in the case of violations of Mauchly s test of sphericity. Partial eta squared ( ) values are reported throughout for effect size of main effects. The alpha 2 p level of statistical significance was set at.05 with the sequential Bonferroni correction 54

55 applied to control for family-wise error where multiple t-test comparisons were conducted. Finally, pairwise comparisons were carried out in the case of significant interactions while Cohen s d is reported for effect size of these comparisons. Results and Discussion Response Accuracy Data Mean scores and standard errors of response accuracy for the skilled and less-skilled groups are presented in Figure 2.2. First, we hypothesised that participants would be more accurate than chance in both the video and animated conditions. As expected, both groups response accuracy scores were significantly higher than would be expected due to chance in both video and animated display conditions when making depth, direction, and combined anticipation judgments (all p <.01). This primary finding demonstrates that tennis players can use contextual information to make more accurate anticipation judgments than would be expected due to chance when no postural information is available from the visual display to inform their judgments. This finding supports previous research showing that contextual information plays an important role in anticipation (Abernethy et al., 2001; Triolet et al., 2013). Moreover, data extend previous work by confirming that anticipation judgments may be made more accurately than would be expected due to chance in the absence of pertinent postural cues. The high mean scores for combined judgments of both skilled and less-skilled groups in the animated condition (49.10% and 44.00% respectively) when compared to chance level (25.00%) indicate that not only is this a useful source of information, but it is a source that both skilled and lessskilled performers are able to exploit. These findings resemble those in studies in which participants have been presented with point-light displays of the opponent alone. In that body of research, performers were shown to be able to use minimal biological motion 55

56 information to anticipate at levels significantly greater than chance (e.g., Abernethy et al., 2008; Abernethy & Zawi, 2007; Ward, et al., 2002). In the present study, the sequential relative movement of the players and the ball flight provided a sufficiently informative basis from which participants were able to anticipate the outcome at better than chance levels. We hypothesised that participants would be more accurate in the video than in the animated condition. Analysis of the response accuracy data was consistent with this hypothesis, revealing a significant multivariate main effect for Display, Wilks Lambda 2 =.06, F(3, 32) = , p <.01, p =.94. The univariate output revealed a significant main effect of Display for the depth, F(1, 34) = , p <.01, 34) = 14.05, p <.01, 2 p =.94, direction, F(1, 2 p =.29, and combined judgment data, F(1, 34) = , p <.01, 2 p =.76. For all judgment measures mean scores were higher in the video condition than in the animated condition. From this it can be inferred that while contextual information enables performers to anticipate well above chance level, postural information makes a significant additional contribution to anticipation skill. Reflecting the above effect sizes, the advantage of viewing video as opposed to the animation was greater for depth response accuracy (normal: 82.19%, animated: 67.38%) than for directional judgments (video: 68.89%, animated: 65.89%). This suggests that postural information may be particularly important for anticipating depth. For example, the most useful cues when anticipating drop shots may emerge from the angle and speed of the racket head on approach to racket-ball contact. Abernethy et al. (2008) showed that when players had to anticipate a smash or drop shot in badminton while viewing a pointlight display showing only the shuttle and either the racket, the arm, the upper body, or 56

57 the lower body, the most informative source at racket-shuttle contact for the skilled players was the racket. We further hypothesised that the increase in accuracy from the animated to the video condition would be more pronounced in the skilled than the less-skilled group, and that skilled participants would be more accurate than their less-skilled counterparts in both video and animated conditions. The MANOVA revealed a significant main effect of 2 Group, Wilks Lambda =.60, F(3, 32) = 7.08, p <.01, p =.40, with the univariate analysis indicating a significant group effect for the depth, F(1, 34) = 7.88, p <.01,.19, direction, F(1, 34) = 11.21, p <.01, = 21.76, p <.01, 2 p = 2 p =.25, and combined judgment data, F(1, 34) 2 p =.39. In each case, skilled participants were more accurate than their less-skilled counterparts. The hypothesised Display Group interaction approached 2 significance overall, Wilks Lambda =.81, F(3, 32) = 2.43, p =.08, p =.19, with the univariate analysis revealing a significant Display Group interaction for direction 2 judgments, F(1, 34) = 7.07, p =.01, p =.17 (see Figure 2.2). To clarify the source of these findings pairwise comparisons revealed significantly higher accuracy for the skilled group for depth (video: p <.01, d = 1.06; animated: p =.02, d = 0.70) and combined (video: p <.01, d = 1.72; animated: p <.01, d = 0.93) judgments across both display conditions, while for direction judgments the group effect was significant and large in the video condition (p <.01, d = 1.36), but smaller and non-significant in the animated condition (p =.07, d = 0.51). Collectively, these findings confirm that skilled players: a) use postural cues picked up from an opponent to anticipate more effectively than their less-skilled counterparts; (Mann et al., 2007) and b) can use contextual information more effectively than less-skilled participants in the absence of postural cues. 57

58 Figure 2.2. Mean (SE) depth, direction and combined response accuracy of skilled and less-skilled participants in video and animated display conditions. *p <.05 To summarise, this study demonstrated that both skilled and less-skilled tennis players are able to use contextual information to anticipate more effectively than would be expected due to chance when postural information is not available. This finding is important because tennis players are often required to anticipate in advance of pertinent postural cues becoming available due to extreme time constraints (Triolet et al., 2013). Furthermore, findings indicate that while both skilled and less-skilled players can use contextual and postural information to anticipate more effectively than would be expected due to chance, both groups can use postural information to anticipate more accurately when it is available than when it is not. Moreover, skilled players can use contextual information to anticipate more effectively than less-skilled players in the absence of postural cues. The findings of this study alone however, are not sufficient to understand how skilled performers process contextual information to anticipate more effectively than their less-skilled counterparts. 58

59 Study 2 In Study 2, we trace the perceptual-cognitive processes underlying anticipation when constrained to anticipate based on contextual information alone. To understand the processes underpinning skilled anticipation performance, researchers have reproduced the conditions of validated representative anticipation tasks, and collected verbal report and gaze data to aid their understanding of how skilled performers perceive and process information differently to less-skilled performers (e.g., McRobert et al., 2011; North et al., 2011; Roca et al., 2011). In anticipation tasks in sports other than tennis, when asked to report their thoughts concurrently or retrospectively, skilled performers have been shown to make more verbal statements which are indicative of elaborate domain-specific memory representations which facilitate the retrieval and utilisation of task-specific information from LTM (e.g., McRobert et al., 2009; North et al., 2011). However, no researchers have investigated the processes underpinning skilled performance when constrained to anticipate based on contextual information alone. It is important to note that researchers have demonstrated that tennis players can verbalise their thoughts during performance providing an indication of the information they are attending to (McPherson, 1999, 2000; McPherson & Kernodle, 2007). This body of work, investigating the cognitive processes underpinning skilled performance in tennis, has shown that during competition, skilled players verbal reports are more detailed and varied than less-skilled players and has provided an indication of the information players attend to during competition. It is also evidence of the utility and effectiveness of verbal reports as a source of data. Williams and Davids (1997) originally collected verbal reports of thoughts in conjunction with eye-movement data to verify the information participants were attending 59

60 to in a soccer anticipation task, finding that the relationship between visual fixation and the information participants verbally reported they were attending to, was dependent on the nature of the task. In a task which evoked the use of foveal vision for information extraction, the area of the display fixated on was also verbally reported by participants, whereas when information was extracted through the use of peripheral vision, participants reported attending to different information to that which they were fixating on during the task. Skilled performers have been shown to exhibit different gaze behaviours to lessskilled performers (e.g., Ward et al., 2002; Williams et al., 2002). While there is a large body of research evidence to indicate that skilled performers use less fixations of longer duration when searching for information in some tasks (Mann et al., 2007), some researchers interested in anticipation in tennis have observed no differences in search rate (numbers of fixations and fixation locations, fixation duration) between high and lowskilled players but instead differences have been demonstrated in the time players spend fixating on various sources of information (Ward et al., 2002; Williams et al., 2002). It has however been recommended that gaze data should be interpreted with caution, as they merely indicate the location of the fovea during task performance, without assessing the potentially important role played by peripheral vision (Mann & Savelsbergh, 2015). In some situations it has been suggested that skilled performers merely adopt an appropriate point upon which to anchor foveal vision. This strategy is thought to accommodate attending to, and processing of task-relevant information picked up from other pertinent areas of the environment through the use of peripheral vision (Ripoll et al., 1995; Vaeyens, Lenoir, Williams, Mazyn, & Philippaerts, 2007; Williams & Elliott, 1999). In such instances, researchers can only make inferences as to what information participants 60

61 attend to, highlighting the need to collect more direct measures such as verbal reports of thoughts in conjunction with gaze data (Williams & Davids, 1998). In this study, we compared skilled and less-skilled tennis players ability to anticipate shot direction and depth when viewing test stimuli in the animated condition, while collecting eye-movement data during trials and immediate retrospective verbal reports of thoughts following selected trials. We used the animated condition as opposed to the video condition because we are interested in the perceptual-cognitive processes underlying anticipation based on contextual information alone, rather than when postural information is available. We hypothesised, based on the findings of Study 1, that both groups would be able to anticipate at levels significantly greater than chance and that the skilled participants would record higher response accuracy scores than their less-skilled counterparts. As skilled participants appear to be able to use contextual information picked up from the display to anticipate more effectively than less-skilled participants, we further hypothesised that the skilled participants in this study would process such information differently to less-skilled participants. Based on LTWM theory (Ericsson & Kintsch, 1995), as well as previous research on anticipation (McRobert et al., 2011; North et al., 2011; Roca et al., 2011), we hypothesised that skilled participants would make more evaluation and prediction statements than less-skilled participants. This finding would be indicative of the skilled participants more elaborate domain-specific memory representations. We further analysed participants verbal reports based on domainspecific keywords to provide an indication of the information participants are consciously attending to when presented only with contextual information. We hypothesised that skilled participants verbal reports would be more detailed and varied than less-skilled participants (McPherson, 1999, 2000; McPherson & Kernodle, 2007), which would be demonstrated by their use of a greater amount of different keywords than less-skilled 61

62 participants. Finally, based on previous research on anticipation in tennis, we hypothesised that the skilled and less-skilled participants search rate would not differ but rather that differences would be observed in the amount of time participants fixate on the various features of the display (Ward et al., 2002; Williams et al., 2002). Methods Participants A total of 10 skilled (Mage = 28.6 years, SD = 4.7) and 10 less-skilled (Mage = 23.7 years, SD = 4.4) male tennis players participated in this study. Six of the skilled participants and four of the less-skilled participants had taken part in Study 1. The time period between the two studies was approximately 10 to 12 months. Skilled participants had a mean of 22.0 (SD = 5.3) years of playing experience, five of whom held an Association of Tennis Professionals (ATP) singles ranking (mean career high ranking of 788 [SD = 559] in the world). Less-skilled participants had a mean of 3.8 (SD = 3.9) years of playing experience. Skilled participants had played tennis for a mean of 14.5 (SD = 4.2) hours per week throughout their career and competed in a mean of 19.2 (SD = 11.9) competitions per year, whereas less-skilled participants had played a mean of 1.6 (SD = 1.5) hours per week, and had not competed in competitive tournaments outside of schools tennis. One participant in the skilled group and two participants in the less-skilled group were left-handed with respect to the hand they normally use to play tennis with, whereas all other participants normally use their right hand. As in Study 1, participants gave their informed consent to take part and were informed that they could withdraw at any time without penalty. The research was carried out according to the ethical guidelines of the lead university. 62

63 Test Stimuli A total of 28 trials in the animated display condition, which were shown to positively discriminate between skilled and less-skilled participants in Study 1, were used as test stimuli. The 28 trials were made up of 8 familiarisation and 20 experimental trials. In Study 1, skilled participants had recorded combined response accuracy scores at least 8.75% more accurate than less-skilled participants for each of the 20 experimental trials used. The number of shots played prior to the opponent s occluded shot in trials was between 2 and 12. Trials lasted between 5 and 18 seconds including a still frame that was presented for three seconds prior to the racket-ball contact of the serve at the beginning of the point. In experimental trials the number of shots played to the four areas of the court on the final occluded shot were: deep-left (8); deep-right (8); short-left (2); and short-right (2). The order of the bounce location of the opponent s final occluded shot was randomised throughout trials. An inter-trial interval of 6 seconds was employed in all trials except the eight trials in which participants were required to provide an immediate retrospective verbal report, with the inter-trial interval being extended until participants had finished reporting their thoughts in those trials. Materials, Apparatus, and Set-Up The experimental set-up and visual angle employed was the same as in Study 1. In addition, participants gaze was recorded using a Mobile Eye eye-tracking system (Applied Science Laboratories, Bedford, MA, USA) during trials. The head-mounted system is integrated in a pair of glasses connected to a digital transmission unit, which is wirelessly connected to the recording device and worn by the participant in a small backpack. The gaze recording system records two images with two separate cameras; an 63

64 image of the participant s eye and the scene image. The system integrates these two images to create one video recording of the scene video with a superimposed gaze cursor. The gaze data were analysed using ASL Results Plus (Applied Science Laboratories, Bedford, MA, USA). The system is accurate to ± 1 visual angle, with a horizontal and vertical precision of 1. Verbal reports of thoughts were recorded using a lapel microphone, a compact diversity receiver, a body-pack transmitter (ew112-p G3; Sennheiser, Wedemark, Germany) and a hand-held recording device (Zoom H5; Zoom Corporation, Tokyo, Japan). Procedure Prior to commencing testing, participants were given a brief overview of the experimental protocol. The microphone was attached to the lapel of the participant and the body-pack transmitter to their belt-strap. Participants then took part in between 20 and 35 minutes verbal report training on how to provide retrospective verbal reports of thoughts to ensure participants provided only level 1 and 2 verbalisations (based on Ericsson & Kirk [2001] using an adaptation of Ericsson & Simon s [1993] protocol). According to Ericsson and Simon (1980), Level 1 verbalisations are reports of heeded information, Level 2 verbalisations are reports of information that was heeded but is in some mode other than verbal, such as visual information. Ericsson and Simon (1980) stated that verbal reports should only be made up of Level 1 and 2 verbalisations as the researcher is interested in information processed during the task as opposed to reports of information that is not heeded during the task (Level 3 verbalisations) and is therefore not reported as a result of the cognitive processes underlying performance. The training consisted of instructions on how to report thoughts retrospectively, description of how 64

65 providing verbal reports differs to normal conversation, and practice providing retrospective verbal reports on generic tasks (for details on training participants to provide valid verbal reports of thoughts, see Eccles, 2012). The training was designed to encourage participants to only report Level 1 and 2 verbalisations through feedback about the verbal reports they gave during these tasks. Participants were encouraged to ask questions throughout the training and were provided with good and bad examples and the difference between such examples, of retrospective verbal reports based on these tasks. Following the verbal report training, participants were fitted with the gaze recording system, placing the eye-tracking glasses on their head and the digital transmission unit in a small backpack on their back. The system was calibrated using seven non-linear calibration points presented as a grid on the visual display which encompassed the entire area of the display participants could potentially fixate on. This procedure was undertaken to ensure that the participants point of gaze was accurately recorded. Calibration of the system was checked prior to starting the familiarisation trials, between familiarisation and experimental trials, and periodically during testing. Following calibration of the gaze recording system, participants viewed eight familiarisation trials from a distance of 5 m. Trials were occluded at the opponent s racket-ball contact, at which point participants responded in the same way as in Study 1. Participants were then asked to provide an immediate retrospective verbal report of their thoughts following each of the familiarisation trials. The same procedure was in place for the 20 experimental trials; however participants only provided a verbal report of their thoughts when asked to, following eight randomly ordered trials (the eight most discriminating trials from Study 1). All participants provided verbal reports on the same eight trials. The testing protocol was completed within 60 minutes. 65

66 Data Analysis Response accuracy data The number of correct trials out of twenty was expressed as a percentage with respect to depth, direction, and combined judgments. One-sample t-tests were conducted to compare response accuracy scores of the two groups with what would be expected due to chance. Independent t-tests were conducted to determine differences in response accuracy between groups with percentage of correct depth, direction, and combined judgments serving as dependent variables. Verbal report data The verbal reports provided in the experimental trials were transcribed and coded using Ericsson and Simon s (1993) protocol analysis method, further developed by Ward (2003). Reports were initially segmented using natural speech and other syntactical markers. Reports were then coded as being made up of three statement types: monitoring statements which were coded as all statements in which participants recalled current actions or events; evaluation statements, which were all statements in which participants assessed the situation relative to a specific event; and prediction statements, which were all statements in which anticipation of future or potential future events was evident (Ericsson & Simon, 1993). These three categories were used due to their use in previous studies on anticipation using verbal reports. Moreover, these categories have been successfully employed to interpret the cognitive processes underpinning expertise. Using the same coding system would therefore allow us to compare our findings with those of other related studies. Pairwise comparisons were first conducted to determine whether there was a difference in the length of verbal reports (number of words) made by skilled and less-skilled participants. This procedure was undertaken to ensure that any 66

67 differences were due to the quality and/or type of participants reports as opposed to the length of their reports. Next, to examine the type of verbal statements used by skilled and less-skilled participants, a 2 3 (Group [skilled, less-skilled] Statement Type [monitoring, evaluation, prediction]) ANOVA was conducted, with Group as the between-participant factor and Statement Type as the within-participant factor. Finally, pairwise comparisons were conducted to investigate differences between participants in the type of statement made. To conduct an initial investigation of what sources of contextual information participants were using from the display to inform their judgments a bespoke domainspecific keyword coding/categorisation system was developed. Roca et al. (2013) used a similar approach when identifying interactions between the perceptual-cognitive skills underlying anticipation in soccer. Keywords were broken down into six categories: player positioning keywords which referred to the movement and positioning of the players; court geometry referring to the angles and spaces presented between players; shot type referring to any type of shot used in tennis (e.g., forehand/lob); ball flight referring to the speed, spin, and height of shots hit; shot placement referring to the depth and direction of a player s shot; and balance of power referring to the tactical situation within the rally. Once keywords were coded within reports, the percentage of participants reports containing each of these keywords was calculated. A 2 6 (Group Keyword Type [player positioning, court geometry, shot type, ball flight, shot placement, balance of power]) ANOVA was conducted, with Group as the between-participant factor and Keyword Type as the within-participant factor. 67

68 Gaze data Search rate data were made up of the mean number of fixations, mean number of fixation locations, and mean fixation duration. A fixation was defined as the participant s point of gaze staying stationary for three frames or more (>/= 100 ms) within 1.5 of movement tolerance (see Williams & Davids, 1998). To determine any between-group differences, independent t-tests were conducted for search rate, with mean number of fixations, mean number of fixation locations, and mean fixation duration as the dependent variables and Group as the independent variable. Percentage viewing time was defined as the portion of time the participant spent fixating on a particular area of interest. Areas of interest were: the (receiving) player; the opponent; the ball flight; the near side of the court; the far side of the court; and the play area (any other area of the display within which the ball could potentially travel throughout the trial); and unclassified areas. Unclassified fixations were subsequently omitted from the analysis as they made up less than 1% of total fixation time. Ball flight as opposed to the ball alone was used as an area of interest. The ball flight area of interest subtended a visual angle of 6.0 in the direction the ball was travelling and 1.2 in the corresponding perpendicular plane, thus encompassing when participants were looking at the ball as well as if they made a visual saccade to where the ball was going to bounce or be played to (Croft, Button, & Dicks, 2010; Singer et al., 1998), and if participants gaze lagged behind the ball (Land & McLeod, 2000). The size of the shape used to identify when participants gaze was on ball flight was maintained constant throughout the analysis. To examine the amount of time skilled and less-skilled participants spent viewing the various areas of the visual display, a 2 6 (Group Fixation Location [player, opponent, ball flight, near side, far side, play area]) ANOVA was conducted with 68

69 Group as the between-participant factor and Fixation Location as the within-participant factor. The first author analysed all trials, and conducted further analysis to determine intra-observer reliability one week later. Inter-observer agreement was conducted by an independent investigator. Intra- and inter-observer reliability for verbal report and gaze data ranged from 84% to 93% (see Thomas, Nelson, & Silverman, 2005 for procedures used to determine intra- and inter-observer reliability). The Greenhouse-Geisser correction was employed in the case of violations of Mauchly s test of sphericity. Partial eta squared ( ) values are reported for effect size of main effects and the alpha level of statistical significance for all tests was set at.05 with the sequential Bonferroni correction applied to control for family-wise error where multiple t-test comparisons were conducted. Pairwise comparisons were carried out in the case of significant interactions and main effects. Cohen s d is reported for effect size of these comparisons. 2 p Results and Discussion Response Accuracy Data We hypothesised, based on the findings of Study 1, that both skilled and less-skilled participants would anticipate at levels significantly greater than chance, and that the skilled participants judgments would be more accurate than the less-skilled participants. Mean percentage scores and standard errors for response accuracy of both skilled and less-skilled groups are presented in Figure 2.3. Both skilled and less-skilled participants response accuracy scores for depth, direction, and combined judgments were significantly higher than would be expected due to chance (p <.01). This finding reaffirms the ability of tennis players to use contextual information to anticipate effectively when postural 69

70 information is not available, signifying the important role this form of information plays in anticipation (Abernethy et al., 2001; Triolet et al., 2013). This finding is particularly important due to the extreme time constraints experienced during fast ball sports such as tennis which mean it may not always be feasible for players to wait for pertinent postural cues to become available. Skilled participants were significantly more accurate than lessskilled participants when making depth (p <.01, d = 1.59) and combined (p <.01, d = 1.86) but not direction (p =.20, d = 0.32) judgments. Figure 2.3. Mean (SE) depth, direction and combined response accuracy across groups in the animated display condition. *p <.05 Verbal Report Data Statement Type First, pairwise comparisons revealed no significant difference between the number of words contained in skilled and less-skilled participants verbal reports (p =.28, d = 0.27). Any differences that may be observed between or within groups are therefore a result of the quality or type of report provided as opposed to the report length. We 70

71 hypothesised, based on LTWM theory (Ericsson & Kintsch, 1995) and previous research on anticipation (McRobert et al., 2011; North et al., 2011; Roca et al., 2011), that skilled participants would make more evaluation and prediction statements than their less-skilled counterparts. Mean number of statements and standard errors are presented in Figure 2.4. There was no significant Group Statement Type interaction, F(1.22,22.01) =.60, p = 2.48, p =.03, however a main effect of Statement Type was observed, F(1.22,22.01) = 4.60, p =.04, =.20. No main effect of Group was observed, F(1,18) =.53, p =.48, 2 p =.03. Pairwise comparisons for Statement Type revealed that participants made significantly more monitoring (M = 10.10, SE = 2.09) and prediction (M = 6.6, SE =.54) than evaluation (M = 4.25, SE =.68) statements (p =.04, d = 0.86, and p =.02, d = 0.81 respectively). To test our a priori prediction that the skilled group would use more evaluation and prediction statements than the less-skilled group, we conducted two planned contrasts. In support of Ericsson and Kintsch s (1995) LTWM theory, the first of these revealed significant differences between skilled (M = 5.60, SE =.87) and lessskilled groups (M = 2.9, SE = 1.04) for evaluation statements (p =.03, d = 0.89), while the difference between skilled (M = 7.4, SE =.52) and less-skilled (M = 5.8, SE =.94) groups for prediction statements was not significant, (p =.08, d = 0.66). Findings indicate that when presented with only contextual information, skilled participants appear to evaluate the situation and potential outcomes more effectively than their less-skilled counterparts. This effect potentially contributes to their more accurate anticipation judgments when constrained to anticipate based on contextual information alone. 2 p Our findings are in support of previous research which has shown that skilled performers make more statements which are indicative of more elaborate memory representations, that is more evaluation statements. Although the difference between 71

72 groups in the amount of prediction statements made did not reach statistical significance, the observed effect size was medium to large. We therefore suggest that similar cognitive processes underlie skilled anticipation when constrained to use contextual information alone as when postural information is readily available for processing (McRobert et al., 2009; North et al., 2011; Roca et al., 2011). Skilled participants are likely to have regularly encountered similar situations and as such may have developed elaborate memory representations from which they can access information for evaluative purposes. When skilled participants perceive contextual information in the form of sequential relative movements of the players and the ball flight in the lead up to the event, it is possible that some feature(s) picked up from the display, may act as a retrieval cue, which is activated to retrieve task-relevant information from LTM. This information potentially helps guide participants judgments by allowing for more thorough evaluation of the event, and prediction of potential event outcomes, relative to previously experienced situations. While previous research on anticipation has always presented participants with postural information which could act as the retrieval cue needed to anticipate more effectively, our findings indicate that skilled performers may be able to use some form of contextual information as a retrieval cue to access pertinent task-relevant information from LTM during anticipation. 72

73 Figure 2.4. Mean (SE) number of verbal statements made by skilled and less-skilled participants. *p <.05 Keyword Type Based on the research of McPherson and colleagues (McPherson, 1999, 2000; McPherson & Kernodle, 2007), we hypothesised that skilled participants verbal reports would be more detailed and varied than less-skilled participants. To analyse the Keyword Type data we first calculated the percentage of participants reports containing each type of keyword. These data are presented in Figure 2.5. ANOVA revealed the Group Keyword Type interaction did not reach statistical significance, F(2.92,52.62) = 2.46, p =.07, =.12, however significant main effects of Group, F(1,18) = 5.99, p =.03, 2 p 2 p = 2.25 and Keyword Type, F(2.92,52.62) = 19.56, p <.01, p =.52, were observed. The main effect of Group indicates that, in line with our hypothesis, skilled participants used a greater number of different keywords per report (M = 2.73, SE =.22) than less-skilled participants (M = 1.94, SE =.24). 73

74 Further independent t-tests revealed that significantly more of the skilled participants reports contained court geometry (p <.01, d = 1.28) and shot type (p <.01, d = 1.30) keywords. This finding provides an indication of the sources of information consciously attended to by skilled participants when constrained to anticipate based on contextual information alone, as would be the case under extreme time constraints, providing a novel contribution to the research literature. When less-skilled participants were making statements containing player positioning keywords such as I shot to the left hand side of my opponent, and I moved towards the net, in the same trial skilled participants were making statements containing court geometry and shot type keywords such as The space was on the left hand side for an angled passing shot. It appears that the skilled participants are better able to use domain-specific information picked up from the relative movement and positioning of the two players to form meaningful evaluations about the availability of spaces and angles between players and the resultant potential shot selections of the opponent, as opposed to merely monitoring the players positioning on the court. Furthermore, while the type of shot being played (e.g., forehand lob) is not immediately apparent from the visual display as the players bodies and rackets are not visible, skilled participants appear to infer the type of shot being hit based on the information presented. Knowledge of the type of shot usually hit from a particular position on the court may allow the skilled players to more effectively evaluate the situation and interpret the tactical intentions of the opponent. 74

75 Figure 2.5. Mean (SE) percentage of reports containing keywords for skilled and lessskilled participants. *p <.05 Gaze Data Search Rate We hypothesised based on previous research on anticipation in tennis (e.g., Ward et al., 2002; Williams et al., 2002), that no differences would be observed in search rate between groups and the differences would, in contrast, exist in the amount of time participants spent fixating on particular areas of the display. Pairwise comparisons revealed that skilled participants fixated on significantly fewer of the six potential fixation locations per trial than less-skilled participants (M = 4.15, SE =.13 vs. M = 4.67, SE =.20, p =.02, d = 1.04). No differences were observed between groups for mean number of fixations or mean fixation duration. The data showing that skilled participants fixated on significantly fewer areas of the display per trial potentially indicate that skilled participants employed a more selective and, as such, more efficient gaze behaviour than less-skilled participants (Williams & Davids, 1998). However, it must be noted that while the effect size of this difference is large, the between-groups difference is small 75

76 (approximately half of a fixation location per trial). We therefore draw conclusions from these data tentatively. Percentage Viewing Time Mean percentage viewing time and standard errors are presented in Figure 2.6. ANOVA revealed a significant Group Fixation Location interaction, F(3.10,49.54) = 3.98, p =.01, 2 p =.20, and a main effect of Fixation Location, F(3.10,49.54) = 17.37, p 2 <.01, p =.52. Bonferroni-adjusted pairwise comparisons revealed that participants fixated the ball flight more than any other area. As hypothesised, differences were observed between groups for the amount of time spent fixating on the various features of the display. Pairwise comparisons revealed that skilled participants fixated Ball Flight for a greater percentage of time (M = 42.55%, SE = 3.63 vs. M = 24.07%, SE = 4.58, p <.01, d = 1.50) than less-skilled participants. It is important to reiterate that we can only infer from gaze data the information participants may be processing from the visual display. It is possible that the skilled participants fixated for a greater amount of time on ball flight because they perceived this to be the most useful form of information available. However, based on previous findings (e.g., Ripoll et al., 1995; Vaeyens et al, 2007), it is possible that they are strategically anchoring gaze on the ball flight to more effectively extract and process pertinent information about the movement of the players relative to each other and the ball. Anchoring gaze on the ball flight may have allowed the skilled participants to more effectively extract informative cues arising from the relative movement and subsequent positions of the players through the use of peripheral vision. While skilled participants fixated on ball flight for a greater amount of time than any other area of the display, this attention to ball flight information is not evident in their verbal reports, in which ball flight is referred to comparatively little. Although skilled participants are 76

77 fixating for the longest amount of time on ball flight in this task, they may be concurrently attending to other, more important information from the display (cf., Williams & Davids, 1997). These data, in conjunction with the verbal report data, provide an initial indication of the sources of contextual information skilled tennis players fixate on and attend to when anticipating based on contextual information alone. Figure 2.6. Mean (SE) percentage viewing time relative to fixation location of skilled and less-skilled participants. *p <.05 To summarise, this study reaffirmed the importance of contextual information in anticipation. We demonstrated that skilled tennis players process contextual information differently to, and in essence, more effectively than less-skilled players. In support of LTWM theory (Ericsson & Kintsch, 1995), skilled participants were shown to make more evaluation statements which are indicative of skilled participants more elaborate domainspecific memory representations. The verbal reports of skilled participants were more detailed and varied relative to domain-specific keywords when compared with lessskilled counterparts. Furthermore, gaze behaviours differed between skill groups, with skilled participants fixating on the ball for greater amounts of time than the less-skilled participants. We tentatively proposed that skilled participants employed a visual 77

78 anchoring strategy to more effectively extract pertinent information from the display through the use of peripheral vision. General Discussion We examined the ability of skilled and less-skilled tennis players to anticipate the depth and direction of an opponent s shot when postural information was either retained or omitted from the visual display, such that participants were constrained to anticipate based on the sequential relative movements of the players and the ball flight alone, or this information as well as postural information. We provided a novel contribution, investigating whether or not participants could anticipate effectively based on contextual information alone, as would be the case when constrained to anticipate in advance of pertinent postural cues becoming available due to extreme time constraints. We hypothesised, based on previous research (Abernethy et al., 2001; Triolet et al., 2013), that participants would be able to use this form of contextual information to anticipate effectively, and that the findings would indicate the extent to which they could do so relative to when postural information was also available. Moreover, we expected skilled participants to be able to use this information more effectively than their less-skilled counterparts (Abernethy et al., 2001). We further aimed to contribute to the development of psychological theory by investigating the perceptual-cognitive processes underlying anticipation when constrained to anticipate based on contextual information alone. We hypothesised, based on LTWM theory (Ericsson & Kintsch, 1995), that skilled participants would make more evaluation and prediction statements which would be indicative of more elaborate domain-specific memory representations and that skilled participants gaze behaviour would differ to that of less-skilled participants, in relation to the amount of time spent viewing the various features of the display (Ward et al., 2002; Williams et al., 2002). 78

79 First, both groups of tennis players were able to anticipate at levels greater than chance when constrained to anticipate based on contextual information alone, with skilled and less-skilled groups recording a mean combined judgment score of 60.30% and 47.00% over the two studies respectively. These high response accuracy scores provide a clear indication of the importance of this form of contextual information in anticipation. This finding is novel since previously researchers have only been able to infer that contextual information may be used in isolation to anticipate effectively since postural information has always been available to participants (e.g., Abernethy et al., 2001; Triolet et al., 2013). We report the first attempt to remove postural information from the display such that the extent to which performers can use contextual information to anticipate could be more directly examined. Second, our findings illustrate that while contextual information in the form of the sequential relative movement of the players and the ball flight is an important source of information that can be used to anticipate effectively by skilled and less-skilled performers, the accuracy of participants anticipation judgments increases substantially when they are also provided with postural information from the opponent, revealing an additive effect of both sources of information. Loffing et al. (2011) found an effect in the opposite direction (i.e., that when tennis players were provided with contextual information in the form of court positioning in addition to postural information, the accuracy of their judgments increased in comparison with when postural information was presented in the absence of contextual information). It is likely that the information upon which performers base their anticipation judgments in a real-world task is situation dependent (Roca et al., 2013) and that they conceivably opt for the most reliable source of information available. The findings of Triolet et al. (2013) indicate that in professional tennis, when players feel anticipation is necessary to avoid losing the point, they may 79

80 only be able to wait for pertinent postural cues to become available on a small number of instances due to the extreme time constraints involved. It is therefore likely, and it seems strategically viable, that players continuously process contextual information to inform potential anticipation judgments throughout rallies, acting on these advanced judgments in the case of severe time constraints, and calling upon more reliable postural information from the opponent, if the time taken to reach the opponent s shot is predicted to be less than it would take for the ball to pass the player. Third, we have demonstrated how skilled tennis players more effectively process contextual information to anticipate with greater accuracy than their less-skilled counterparts. Skilled players appear to more effectively extract pertinent contextual information picked up from the visual display. This information may act as a retrieval cue, which is activated to retrieve task-relevant information from LTM, resulting in more effective evaluation of the presented information and potential outcomes of the opponent s upcoming shot. While these data provide an indication of how skilled performers process contextual information in the absence of postural information, LTWM theory predicts that when task-relevant options exist, skilled performers would be expected to generate a greater number of task-relevant options than less-skilled performers (Ward et al., 2013). An interesting line of research would therefore be to investigate how contextual and postural information interact with respect to option generation, as this has the potential to provide useful information for the design of testing and training protocols. It is possible though, that in the competitive environment skilled performers would generate more task-relevant options than less-skilled performers far in advance of the event based on the contextual information available (information that in absence of pertinent postural cues can be used to anticipate effectively), and that the 80

81 perceived feasible options would become fewer but more reliable in the build-up to the event due to the availability of pertinent postural cues. Although we have found that contextual information picked up in the form of the sequential relative movements of the players and the ball flight in the lead-up to the event, along with the resulting court positioning of the players at the moment of the event is a useful source of information which can be exploited by both skilled and less-skilled tennis players, we do not yet know what information specifically participants use to inform their judgments. For example, the gaze data indicated that skilled participants fixated for longer on the ball flight than their less-skilled counterparts, whereas the verbal report data indicated that skilled participants attended more to court geometry and shot type than their less-skilled counterparts. While it is possible that the ball flight is the most important and informative source of information available from the display, it is possible the relative positioning of the two players at the moment of racket-ball contact may provide the most useful information. In a similar way to how researchers manipulated point-light displays to determine the most informative sources of biological motion information picked up from an opponent (e.g., Cañal-Bruland, van Ginneken, van der Meer, & Williams, 2011; Huys et al., 2009), there is a need for further research which manipulates the visual display to facilitate evaluation of the role and importance of the various sources of contextual information contained within rallies. Such an investigation would not only enhance our knowledge of perceptual-cognitive expertise but enable recommendations to be made as to what information developing players should attend to in anticipation training programs. Moreover, while researchers have demonstrated the effectiveness of anticipation training programs in both laboratory and field-based settings (Abernethy, Schorer, Jackson, & Hagemann, 2012; Williams et al., 2002), only one study to our 81

82 knowledge has trained participants to use contextual information when anticipating (Mann et al., 2014). In recent years there has been a shift towards, and a call for, more research investigating the multidimensional nature of anticipation (Cañal-Bruland & Mann, 2015). For example, Roca et al. (2013) provided an illustration of how the perceptual-cognitive skills and processes underlying anticipation interact as a function of the unique task constraints placed on the performer. The number of research studies investigating how contextual information contributes to anticipation is increasing (e.g., Crognier & Féry, 2005; McRobert et al., 2011) and the work of Triolet et al. (2013) has provided us with behavioural evidence of the use of contextual information in a real-world setting. The present paper not only adds to a growing body of work in this area, but through the use of a novel methodological approach, and multiple process-tracing measures has been able to determine with much more certainty, if and how, performers utilise and process contextual information to inform their anticipation judgments. The viewing perspective employed in this study may be considered to be a limitation in the study design as footage does not replicate the first person viewing perspective normally experienced in tennis. However, to examine the importance of contextual information presented as the sequential relative movement of players and the ball flight in the lead up to the event, it was necessary for participants to be able to see both players and how the velocities of their movements and positions on the court changed relative to the ball and each other throughout the rally. Additionally, while the third person perspective used is not that which would be experienced in a match situation, the relevant information presented is the same as that which would be available in a match (e.g., player positioning/velocity of movement/previous shots hit) albeit in a different form. Moreover, we report significant differences in anticipation across skill groups 82

83 providing construct validity for the approach. On a similar point, while our gaze behaviour data may not exactly replicate those which would be reported in an on-court setting (e.g., Mann, Farrow, Shuttleworth, & Hopwood, 2009), our findings suggest that skilled participants employed more effective gaze behaviours than their less-skilled counterparts. Furthermore, participants fixated on the player on the near side of the screen a mere 5.15% of total viewing time and only mentioned the near player approximately once per trial in their verbal reports, meaning that only a small amount of time was spent fixating on and attending to the one element of the display that would not be visible in a real-world scenario. Due to technological limitations, there were differences between the video and animated display condition other than the mere omission of postural information in the animated condition. Other information such as the environmental surroundings, the crowd viewing the match, and the umpire were present in the video condition but not in the animated condition and we therefore cannot assume that the higher response accuracy scores reported in the video condition in comparison with the animated condition are solely a result of the additional postural information that participants were presented with. However, based on anticipation research to date (for a review, see Mann et al., 2007) we considered postural information to be the source of information with the most potential to inform participants anticipation judgments that was present in the video but not the animated condition. A further limitation of the present study is the relatively small sample size that compromises the statistical power of our analyses. The difficulty associated with recruiting very highly-skilled athletes, particularly in individual as opposed to team sports, is a common issue faced by researchers in this area. In practice it means that only larger effect sizes will attain statistical significance. In the studies presented here, we were 83

84 able to use the same number or more participants than in several related studies (e.g., Abernethy et al., 2001; Farrow & Reid, 2012; McRobert et al., 2011) while being fewer than in others (e.g., Loffing et al., 2011; Loffing & Hagemann, 2014). Accordingly, additional research is needed to determine whether comparisons that resulted in small or medium effect sizes are replicated, thereby allowing us to more confidently infer associated small or medium effects in the population. Conversely, statistically significant findings in the present study reveal large effects in comparisons between high-skilled players and less-skilled but experienced players rather than novices. They are therefore good candidates for having practical as well as statistical significance. In conclusion, we used a novel experimental design to demonstrate that contextual information picked up from the sequential relative movement of the players and the ball flight in the lead up to the event effectively informs anticipation and that these judgments are made more accurate upon presentation of postural information. Due to the extreme time constraints involved in tennis, players may not always be afforded the opportunity to wait for pertinent postural cues to become available. As such the ability to accurately anticipate based on contextual information in advance of such cues may prove to be a key contributing factor to expert performance in time-constrained domains. Furthermore, we found that skilled tennis players employed different gaze behaviours and more thoroughly evaluated this form of contextual information to anticipate more accurately than their less skilled counterparts. Finally, effective use of contextual information may not only set the scene for accurate anticipation when pertinent postural cues becomes available, but crucially it may allow for highly accurate anticipation when extreme time constraints mean that waiting several moments longer is not a feasible option. 84

85 Chapter 3 The sources of contextual information contributing to skilled anticipation 85

86 Abstract In dynamic, temporally-constrained tasks, individuals often need to anticipate what will happen next prior to information becoming available within the environment. In such situations, the availability of contextual information can facilitate anticipation. While many researchers have identified the specific sources of postural information facilitating anticipation, comparatively few have investigated the specific sources of contextual information that do so. In two studies, we presented skilled and less-skilled tennis players with animations of rallies from real matches that omitted access to postural information from the opponent, constraining participants to anticipate based on contextual information alone. In Study 3, participants anticipated the outcome of an opponent s shot in three conditions in which the sequence length preceding the same occluded shot was varied. Participants anticipated shot direction more accurately when the preceding shot sequence was presented than not. In Study 4, we presented animations that depicted the ball, the players, or both, in either dynamic or still form. Those conditions in which only the ball was depicted yielded the lowest direction response accuracy scores. It appears that both player and ball motion are required to provide the context under which skilled performers can consciously pick up and utilise information to anticipate more accurately than less-skilled counterparts. 86

87 Introduction At the highest levels of performance in dynamic, temporally-constrained domains, such as military combat, sport, and law enforcement, performers often have to make quick and accurate judgments based on minimal information. It is well established that skilled performers extract and utilise postural information to anticipate the opponent s intentions more effectively than their less-skilled counterparts (e.g., Abernethy & Russell, 1987; Williams & Burwitz, 1993). Additionally, researchers have identified the specific sources of postural information used to do so, for example, the hips in the soccer penalty kick (Causer et al., 2017), or the arm and racket in the tennis serve (Jackson & Mogan, 2007). In fast ball sports, the time taken to process and respond to ball flight information often exceeds the time it takes for the ball to pass the receiving player (Singer, 2000; Williams, Davids, & Williams, 1999), and under particularly extreme time constraints, the same may even be true when responding based on pertinent postural cues picked up some brief moments prior to event occurrence (Triolet et al., 2013). When skilled performers are not afforded the time to wait for pertinent postural cues to become available from the opponent s emerging movement pattern, they are thought to rely on contextual information to anticipate the opponent s intentions (Abernethy et al., 2001; Müller & Abernethy, 2012). In this paper, we focus on determining the extent to which performers can use different sources of contextual information (shot sequencing, player and ball motion and positioning) to anticipate independent of pertinent postural cues. Buckolz et al. (1988) suggested that two types of information can be used to anticipate the intentions of an opponent; postural and contextual information. Postural information arises from the movement pattern of the opponent, whereas contextual information is available prior to pertinent postural cues becoming available. Moreover, contextual information remains available as the movement pattern of the opponent 87

88 develops. Buckolz et al. originally suggested that contextual information exists in numerous forms such as knowledge of the opponent s strengths and weaknesses, climatic conditions, and the relative positioning of the players on the playing terrain. In recent years, researchers have begun to investigate how certain sources of contextual information such as score-dependent patterns of play (Farrow & Reid, 2012), knowledge of an opponent s action tendencies (Navia et al., 2013), or court positioning of the opponent (Loffing & Hagemann, 2014) influence anticipation judgments in skilled performers. These studies have reported that skilled performers in particular, can utilise this information to adjust their expectancies relating to upcoming events. It is thought that as a result of their extended experience within a domain skilled performers develop Long Term Working Memory (LTWM) skills (Ericsson & Kintsch, 1995), which allow rapid access to, and retrieval of, information stored in Long Term Memory (LTM). Skilled performers can encode the presented information and associate it with a retrieval cue in Short Term Memory (STM), which allows access to information in LTM about the relationship between the presented information and potential event outcomes. These retrieval structures allow skilled performers access to task-relevant options, which can be evaluated to inform an accurate judgment rather than merely prescribing a set response (Ericsson et al., 2000). In the aforementioned studies (e.g., Loffing & Hagemann, 2014), skilled performers were shown to adjust their expectancies of potential event outcomes based on contextual information available in addition to emerging postural cues, the suggestion being that their more advanced domain-specific knowledge facilitates integration of pertinent contextual information into the anticipation process. Skilled performers, in particular, appear able to make remarkably accurate anticipation judgments in advance of pertinent postural cues becoming available. For 88

89 example, skilled performers have been shown to accurately anticipate the outcome of an opponent s shot approximately 580 ms and 720 ms prior to the opponent striking the ball in squash (Abernethy et al., 2001) and tennis (Triolet et al., 2013), respectively. Triolet et al. (2013) observed that highly skilled tennis players often began to respond to the opponent s shot early because they were placed under extreme time constraints, for example, when the opponent was attacking from inside the court. Abernethy et al. (2001) suggested that these accurate anticipatory movements were likely to be due to the use of contextual information picked up from sources such as the preceding shot sequence. Murphy et al. (2016) sought to clarify whether contextual information can facilitate anticipation independent of postural information. As well as viewing video footage of rallies from real tennis matches, skilled and less-skilled tennis players viewed animations of the same rallies, in which each of the players were replaced by a cylinder and their rackets were not visible. When viewing the animations, participants were constrained to anticipate based on contextual information alone, eliminating the possibility that their judgments could be based on very early occurring postural cues. Although participants anticipated more accurately when viewing the video footage, both groups were more accurate than chance when viewing the animations, with the skilled participants being most accurate. This is further indication (see Abernethy et al., 2001; Triolet et al., 2013) that when skilled performers need to anticipate in advance of pertinent postural cues becoming available, they can draw upon contextual information to do so. Murphy et al. suggested that even when constrained to anticipate based on contextual information alone, skilled participants were able to access associated task-relevant information from LTM, which facilitated more accurate anticipation than less-skilled participants. Skilled racket sports players have been shown to play specific shots in response to certain shot sequences (McGarry & Franks, 1996). In each of the research studies 89

90 conducted by Abernethy et al., (2001), Triolet et al. (2013), and Murphy et al. (2016) participants had access to contextual information that could, potentially, be picked up from the shot sequence preceding the to-be-anticipated shot, with the sequence length varying from trial to trial. This approach is in stark contrast to most research on anticipation to date, in which participants have either been presented with isolated actions such that information about the preceding sequence of events is unavailable (e.g., the serve in tennis [Jackson & Mogan, 2007]; the bowl in cricket [Müller et al., 2010]; the penalty kick in soccer [Savelsbergh et al., 2002]) or with sequences that have been presented for a fixed period of time or number of events (e.g., in basketball [Ryu, Abernethy, Mann, Poolton, & Gorman, 2013]; in football [Roca et al., 2011]). In such experimental set-ups, the influence that the preceding sequence of events may have on anticipation is controlled or ignored. Only a few researchers have attempted to determine how having access to the preceding sequence of events influences performance. Gray (2002a, 2002b) used a simulated baseball batting task to demonstrate that the expectancies of college level baseball batters are influenced by the preceding sequence of pitches. For example, when three fast pitches were followed by a fast pitch, their batting was more accurate than if the three fast pitches were followed by a slow pitch. Loffing et al. (2015) presented skilled and novice volleyball players with sequences of four attacking shots which had been manipulated to either always present smashes, lobs, or an alternating pattern of the two shots prior to anticipating a shot which was either congruent or incongruent with the preceding sequential pattern of shots played. Their findings suggested that expectations of action outcomes were more strongly affected by these sequences for skilled compared with less-skilled participants. It has also been demonstrated that skilled performers are better than less-skilled performers at picking up repeated patterns in an opponent s game to facilitate anticipation 90

91 (Farrow & Reid, 2012; Milazzo et al., 2015). McRobert et al. (2011) presented skilled and less-skilled cricket batters with videos of six fast and slow bowlers presented in a random order or videos of four fast bowlers presented in blocks, such that they viewed each of the four bowlers for six bowls in a row. Performance levels were higher when viewing the same bowler over repeated trials. The authors observed that skilled batters adapted their gaze behaviour when repeatedly viewing the same bowler. They suggested that the contextual information provided by the preceding actions of the bowler allowed the skilled performers to adapt their gaze behaviour to pick up information from relevant locations more efficiently than when this information was not available. While this body of work suggests that the sequence of events preceding a critical event influences anticipation, no published research to our knowledge has investigated whether the strength of this influence is affected by the number of events in the preceding sequence. In addition to the preceding sequence of events, there is evidence to suggest that other sources of contextual information such as the positioning of the opponent prior to executing a skill can influence anticipation (Loffing & Hagemann, 2014). To identify the specific sources of postural information that facilitate anticipation, researchers have employed the spatial occlusion paradigm by artificially removing selected parts of the opponent s body/equipment such that their relevance as visual cues can be assessed using video simulations (e.g., Abernethy, 1990; Jackson & Mogan, 2007; Williams & Davids, 1998) and point-light or stick figure displays (e.g., Abernethy & Zawi, 2007; Abernethy et al., 2008; Huys et al., 2009). If there is a significant deterioration in accuracy when one part of the opponent s body or equipment is occluded, researchers can infer that this element is an important cue for accurate anticipation (e.g., the arm and racket in racket sports, Abernethy & Zawi, 2007; Abernethy et al., 2008; Shim, Carlton, & Kwon, 2006). 91

92 Similarly, some researchers have manipulated the information they present to participants to determine the sources of information facilitating expert pattern recognition and recall (e.g., Gorman, Abernethy, & Farrow, 2011; North et al., 2011; Williams, et al., 2012). In two experiments, Williams et al. (2012) manipulated videos of sequences from soccer matches such that they were presented in either still or dynamic format, and then presented videos in which central or peripheral elements were omitted from the dynamic display. They reported a decrement in pattern recognition for skilled soccer players in the still relative to the dynamic condition (see also Sebanz and Shiffrar [2009] for a similar finding in a deception detection task) and subsequently, when central elements were omitted from the display. Williams et al. (2012) suggested a key mechanism underpinning skilled pattern recognition in football is the effective extraction of motion information and that only the relative motions of a few key features (e.g., central midfielders and offensive players) is necessary for effective recognition of domain-specific patterns. However, some contradictory findings have been reported in pattern recall and decisionmaking tasks (Gorman et al., 2011; Gorman et al., 2013). Gorman et al., (2013) reported that pattern recall error was lower and decision making accuracy higher when viewing a still compared to a dynamic display involving basketball plays. This body of research is pertinent because it has been suggested that the ability to recognise and/or recall domainspecific patterns may act as an important precursor to effective anticipation (Cañal- Bruland & Williams, 2010; Gorman et al., 2012). What is particularly clear, however, is that the manipulation of representative test stimuli, is a useful way to determine the sources of information used by performers to make accurate judgments, which could ultimately lead to the identification of underlying mechanisms. To date, those attempting to identify the sources of contextual information that contribute to skilled anticipation (e.g., Loffing & Hagemann, 2014; McRobert et al., 92

93 2011) have done so with an emphasis on determining how this information interacts with postural cues. However, when placed under extreme time constraints, the time it takes for the action to be carried out may be so minimal that waiting for pertinent postural cues to become available is not feasible (Triolet et al., 2013). In such situations performers would be constrained to rely on contextual information alone to anticipate effectively. In this paper, we aim to provide a novel contribution by determining the extent to which specific sources of contextual information facilitate accurate anticipation independent of pertinent postural cues. Study 3 Crognier and Féry (2005) examined whether tennis players anticipate more effectively in situations in which they can impose their game on the opponent compared to situations in which they have fewer possibilities to do so. To examine the effect of imposing one s game on an opponent (referred to as tactical initiative), experienced tennis players played points against an opponent in three conditions involving increasing levels of tactical initiative. The experiment was set up in such a way that sequence length was shortest in the low tactical initiative condition and longest in the high tactical initiative condition. Although the participants anticipated the direction of the opponent s shot most accurately following the longest sequences, because the opponent playing the passing shot was instructed to allow the participants to impose their game on him during these sequences, it is impossible to draw conclusions about whether this increase in accuracy was due to the additional contextual information picked up from the preceding shot sequences. In this study, we examined whether contextual information can be picked up from the sequence of shots played prior to a critical event to facilitate anticipation. A secondary 93

94 aim focused on determining whether the ability to use this information develops with increasing skill level. Finally, we investigated whether the length of the preceding sequence affects anticipation. We presented skilled and less-skilled tennis players with animations of sequences of shots played by players in real matches. In these animations, the bodies of the players were replaced by a cylinder and rackets were not visible (see Murphy et al., 2016), such that the observer was constrained to anticipate based on contextual information alone. To determine whether the preceding shot sequence provides contextual information which can be used to facilitate anticipation, we compared the ability of skilled and less-skilled participants to anticipate the intentions of an opponent when viewing trials that presented one, three or five shots prior to the same shot played by the opponent, occluded at racket-ball contact. First, we hypothesised, based on the findings of Murphy et al. (2016), that for both groups, response accuracy would be significantly higher than chance. Second, based on the same research, we expected skilled participants to be more accurate than less-skilled participants. Third, to our knowledge, the only research that has provided any indication as to whether the sequence length preceding a critical event affects anticipation is that conducted by Crognier and Féry (2005). This aspect of the study was therefore exploratory in nature. Fourth, based on the findings of several previous research studies demonstrating that the judgments of skilled performers are affected by the presentation of contextual information over and beyond the presentation of postural information (e.g., Crognier & Féry, 2005; Farrow & Reid, 2012; Loffing & Hagemann, 2014), we hypothesised that skilled participants would anticipate more accurately when presented with the preceding shot sequence (i.e., three or five shot conditions) in addition to viewing the final occluded shot. We expected that the superior LTWM skills (Ericsson & Kintsch, 1995) possessed by the skilled participants would allow them to access potential relevant 94

95 alternatives to the presented contextual information, allowing them to adjust their expectancies to make more accurate judgments than when this additional information was not presented. Conversely, we expected the accuracy of the less-skilled participants to be unaffected by the presentation of the preceding shot sequence. Methods Participants Altogether, 12 skilled (Mage = 27.1, SD = 4.7) and 12 less-skilled (Mage = 24.7, SD = 5.3) male tennis players participated. Skilled and less-skilled participants had a mean of 20.2 (SD = 4.6) and 4.3 (SD = 3.6) years of tennis playing experience, respectively. Skilled participants held British Tennis ratings of 1.1 to 4.1, whereas less-skilled participants held ratings of 10.2 or did not hold a rating. British Tennis ratings range from 1.1 (highest) to 10.2 (lowest). Skilled participants reported having played a mean of 54.1 (SD = 24.1) matches per year, whereas less skilled players did not play competitively. One participant in the less-skilled group was left-handed and the rest were right-handed players. All participants had normal or corrected vision. Those with corrected vision wore glasses or contact lenses while participating. The research was carried out in line with the lead university s research ethics guidelines. Participants provided informed consent prior to taking part and knew that they could withdraw from testing at any time without consequence. Test Stimuli Player movement and ball trajectory data (Hawk-Eye Innovations Ltd., Basingstoke, UK) from professional tennis matches played at the AEGON Championships (2013) were used to create the test stimuli. The data were input into a rendering engine to generate animations of rallies which could then be viewed on VLC 95

96 media player (VideoLAN, Paris, France). The purpose of using animations as opposed to video footage of rallies was to remove postural information from the players and the resulting racket movements such that participants would be constrained to anticipate based on contextual information alone. Any differences in accuracy between experimental conditions could therefore be attributed to the use of this information (for more details, see Murphy et al., 2016). The test stimuli were animations of real tennis rallies which displayed two players moving around the court playing a point, but which omitted the players bodies and rackets such that they were depicted as a blue and a red cylinder and their rackets were not visible, while the ball was depicted as a yellow dot (see animated display condition in Figure 2.1). Pinnacle Studio 15 editing software (Pinnacle, Ottawa, Canada) was used to edit the animations to occlude at the opponent s racket-ball contact. The criteria used for selecting the shot on which to occlude the footage was the same as that used by Murphy et al. (2016). Shots in which the receiving player was placed under extreme time constraints were selected from the database of rallies. In such conditions (e.g., when one player is attacking the other from inside the court or the distance between the two players is smaller than usual), players are constrained to respond earlier than usual to get to, and return, the opponent s shot effectively (Triolet et al., 2013). Furthermore, a minimum of six shots needed to be exchanged between the two players in the rally for it to be suitable for creating the longest sequence length condition (five preceding shots plus the final occluded shot). A total of 44 sequences of six shots were selected based on these criteria. In an effort to ensure that we only used trials in which the shot sequence preceding the final occluded shot presented relevant contextual information, three experienced coaches independently viewed each of the 44 animated sequences that were occluded at 96

97 the opponent s racket-ball contact on the sixth shot. All of the coaches had over 10 years of experience and had coached players of National and/or International level as well as having played at that level. For each of the 44 trials, each coach rated the extent to which the preceding shot sequence would contribute to successfully anticipating the outcome of the final occluded shot on a scale of 1 to 5 with 1 being not at all and 5 extremely. Only sequences in which coaches reported an average rating greater than or equal to 4 were used as test stimuli, yielding a total of 23 trials. A similar procedure has been used in pattern recall and recognition research, where experienced coaches have identified structure in sequences of play to ensure that the footage being used is representative of structured gameplay (Gorman et al., 2012; North et al., 2009). To determine the effect that increasing the number of shots in the sequence preceding the final occluded shot had on anticipation, participants viewed the 23 experimental trials in three conditions. Trials were edited to display one (short sequence length condition), three (medium), or five (long) shots in the sequence preceding the occluded shot, such that they viewed the same final occluded shot three times. The rationale for this approach was that it allowed for reliable comparison of anticipation response accuracy scores relative to three sequence lengths (see Figure 3.1). Altogether, 69 experimental trials were used (three sets of 23). Short, medium and long trials lasted an average of 1.38 (SD =.26), 4.18 (SD =.38), and 6.83 (SD =.55) seconds respectively. These trials were created from data from 10 matches involving 14 right-handed players, in various rounds of the tournament. 97

98 Figure 3.1. Representation of the number of shots played in each Sequence Length condition prior to the same shot occluded at the opponent s racket-ball contact. Materials, Apparatus, and Set-Up Test stimuli were projected on to a m white projector screen (AV Stumpfl, Wallern, Austria) using a NEC PE401H projector (NEC, Tokyo, Japan). Participants held a racket in their hands as if they were about to play a point and stood 5 m from the screen which allowed for a viewing angle of subtended by the opposing player. A similar viewing perspective has been used previously (Loffing et al., 2011; Murphy et al., 2016). Procedure Participants viewed 18 familiarisation and 69 experimental trials. Different rallies were used in the familiarisation trials than the experimental trials. Six sets of randomised experimental trials were created with two participants from each group viewing one of the six sets. Participants viewed three blocks of 23 trials with a one-minute break between blocks. This break was provided to mitigate against boredom or fatigue effects. Each of the 23 trials were presented three times. A yellow circle was displayed on a black screen for 2 seconds prior to the commencement of the trial to indicate where the ball would be once the trial started and a 6 second inter-trial interval was employed. In the experimental 98